Fachveröffentlichungen

We present a novel approach for a rigorous bundle adjustment for omnidirectional and multi-view cameras, which enables an efficient maximum-likelihood estimation with image and scene points at infinity. Multi-camera systems are used to increase the resolution, to combine cameras with different spectral sensitivities (Z/I DMC, Vexcel Ultracam) or - like omnidirectional cameras - to augment the effective aperture angle (Blom Pictometry, Rollei Panoscan Mark III). Additionally multi-camera systems gain in importance for the acquisition of complex 3D structures. For stabilizing camera orientations - especially rotations - one should generally use points at the horizon over long periods of time within the bundle adjustment that classical bundle adjustment programs are not capable of. We use a minimal representation of homogeneous coordinates for image and scene points. Instead of eliminating the scale factor of the homogeneous vectors by Euclidean normalization, we normalize the homogeneous coordinates spherically. This way we can use images of omnidirectional cameras with single-view point like fisheye cameras and scene points, which are far away or at infinity. We demonstrate the feasibility and the potential of our approach on real data taken with a single camera, the stereo camera FinePix Real 3D W3 from Fujifilm and the multi-camera system Ladybug3 from Point Grey.

@INPROCEEDINGS{schneider*12:bundle, author = {Schneider, Johannes and Schindler, Falko and L{\"a}be, Thomas and F{\"o}rstner, Wolfgang}, title = {{Bundle Adjustment for Multi-camera Systems with Points at Infinity}}, booktitle = {ISPRS Annals of Photogrammetry, Remote Sensing and the Spatial Information Sciences; Proc. of 22nd Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, year = {2012}, note = {to appear} }

Our objective is the categorization of the most dominant objects in facade images, like windows, entrances and balconies. In order to execute an image interpretation of complex scenes we need an interaction between low level bottom-up feature detection and highlevel inference from top-down. A top-down approach would use results of a bottom-up detection step as evidence for some high-level inference of scene interpretation. We present a statistically founded object categorization procedure that is suited for bottom-up object detection. Instead of choosing a bag of features in advance and learning models based on these features, it is more natural to learn which features best describe the target object classes. Therefore we learn increasingly complex aggregates of line junctions in image sections from man-made scenes. We present a method for the classification of image sections by using the histogram of diverse types of line aggregates.

@INPROCEEDINGS{wenzel*12:learning, author = {Wenzel, Susanne and F{\" o}rstner, Wolfgang}, title = {Learning a compositional representation for facade object categorization}, booktitle = {ISPRS Annals of Photogrammetry, Remote Sensing and the Spatial Information Sciences; Proc. of 22nd Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, year = {2012}, volume = {I-3}, number = { 2012}, pages = {197-202}, doi = {10.5194/isprsannals-I-3-197-2012} }

Three methods of automatic classification of leaf diseases are described based on high-resolution multispectral stereo images. Leaf diseases are economically important as they can cause a loss of yield. Early and reliable detection of leaf diseases has important practical relevance, especially in the context of precision agriculture for localized treatment with fungicides. We took stereo images of single sugar beet leaves with two cameras (RGB and multispectral) in a laboratory under well controlled illumination conditions. The leaves were either healthy or infected with the leaf spot pathogen Cercospora beticola or the rust fungus Uromyces betae. To fuse information from the two sensors, we generated 3-D models of the leaves. We discuss the potential of two pixelwise methods of classification: k-nearest neighbour and an adaptive Bayes classification with minimum risk assuming a Gaussian mixture model. The medians of pixelwise classification rates achieved in our experiments are 91% for Cercospora beticola and 86% for Uromyces betae. In addition, we investigated the potential of contextual classification with the so called conditional random field method, which seemed to eliminate the typical errors of pixelwise classification.

@ARTICLE{bauer*11:potential, author = {Bauer, S. and Kor{\v c}, F. and F{\" o}rstner, W.}, title = {The potential of automatic methods of classification to identify leaf diseases from multispectral images}, journal = {Precision Agriculture}, year = {2011}, volume = {12}, pages = {361-377}, number = {3} }

We study the set of domain deformations induced on images of three-dimensional scenes by changes of the vantage point. We parametrize such deformations and derive empirical statistics on the parameters, that show a kurtotic behavior similar to that of natural image and range statistics. Such a behavior would suggest that most deformations are locally smooth, and therefore could be captured by simple parametric maps, such as affine ones. However, we show that deformations induced by singularities and occluding boundaries, although rare, are highly salient, thus warranting the development of dedicated descriptors. We therefore illustrate the development of viewpoint invariant descriptors for singularities, as well as for occluding boundaries. We test their performance on scenes where the current state of the art based on affine-invariant region descriptors fail to establish correspondence, highlighting the features and shortcomings of our approach.

@article{dickscheid*11:coding, author = {Dickscheid, T. and Schindler, F. and F\"{o}rstner, W.}, title = {{Coding Images with Local Features}}, journal = {International Journal of Computer Vision}, publisher = {Springer Netherlands}, issn = {0920-5691}, url = {http://dx.doi.org/10.1007/s11263-010-0340-z}, year = {2011}, pages = {154-174}, volume = {94}, issue = {2}, doi = {10.1007/s11263-010-0340-z} }

We propose a surface segmentation method based on Fast Marching Farthest Point Sampling designed for noisy, visually reconstructed point clouds or laser range data. Adjusting the distance metric between neighboring vertices we obtain robust, edge-preserving segmentations based on local curvature. We formulate a cost function given a segmentation in terms of a description length to be minimized. An incremental-decremental segmentation procedure approximates a global optimum of the cost function and prevents from under- as well as strong over-segmentation. We demonstrate the proposed method on various synthetic and real-world data sets.

@inproceedings{Schindler*11:fast, author = {Schindler, Falko and F\"orstner, Wolfgang}, booktitle = {LNCS, Photogrammetric Image Analysis}, title = {{Fast Marching for Robust Surface Segmentation}}, year = {2011} }

We present a novel surface model and reconstruction method for man-made environments that take prior knowledge about topology and geometry into account. The model favors but is not limited to horizontal and vertical planes that are pairwise orthogonal. The reconstruction method does not require one particular class of sensors, as long as a triangulated point cloud is available. It delivers a complete 3D segmentation, parametrization and classification for both surface regions and inter-plane relations. By working on a pre-segmentation we reduce the computational cost and increase robustness to noise and outliers. All reasoning is statistically motivated, based on a few decision variables with meaningful interpretation in measurement space. We demonstrate our reconstruction method for visual reconstructions and laser range data.

@INPROCEEDINGS{schindler*11:classification, author = {Schindler, Falko and F{\"o}rstner, Wolfgang and Frahm, Jan-Michael}, title = {Classification and Reconstruction of Surfaces from Point Clouds of Man-made Objects}, booktitle = {International Conference on Computer Vision, IEEE/ISPRS Workshop on Computer Vision for Remote Sensing of the Environment}, year = {2011}, pages = {257--263}, }

We present a method to perform bundle adjustment using long sequences of digital images from an omnidirectional camera. We use the Ladybug3 camera from PointGrey, which consists of six individual cameras pointing in different directions. There is large overlap between successive images but only a few loop closures provide connections between distant camera positions. We face two challenges: (1) to perform a bundle adjustment with images of an omnidirectional camera and (2) implement outlier detection and estimation of initial parameters for the geometry described above. Our program combines the Ladybug’s individual cameras to a single virtual camera and uses a spherical imaging model within the bundle adjustment, solving problem (1). Outlier detection (2) is done using bundle adjustments with small subsets of images followed by a robust adjustment of all images. Approximate values in our context are taken from an on-board inertial navigation system.

@inproceedings{schmeing*11, author = {Schmeing, Benno, Läbe, Thomas and Förstner, Wolfgang}, title = {Trajectory Reconstruction Using Long Sequences of Digital Images From an Omnidirectional Camera}, year = {2011}, city = {Mainz}, proceeding = {Proceedings of the 31th. DGPF Conference}, }

Wir stellen einen Ansatz für eine strenge Bündelausgleichung für Multikamerasysteme vor. Hierzu verwenden wir eine minimale Repräsentation von homogenen Koordinatenvektoren für eine Maximum-Likelihood-Schätzung. Statt den Skalierungsfaktor von homogenen Vektoren durch Verwendung von euklidischen Größen zu eliminieren, werden die homogenen Koordinaten sphärisch normiert, so dass Bild- und Objektpunkte im Unendlichen repräsentierbar bleiben. Dies ermöglicht auch Bilder omnidirektionaler Kameras mit Einzelblickpunkt, wie Fisheyekameras, und weit entfernte bzw. unendlich ferne Punkte zu behandeln. Speziell Punkte am Horizont können über lange Zeiträume beobachtet werden und liefern somit eine stabile Richtungsinformation. Wir demonstrieren die praktische Umsetzung des Ansatzes anhand einer Bildfolge mit dem Multikamerasystem „Ladybug3“ von Point Grey, welches mit sechs Kameras 80 % der gesamten Sphäre abbildet.

@inproceedings{ schneider*11:buendelausgleichung, author = {Schneider, Johannes and Schindler, Falko and F\"{o}rstner, Wolfgang}, title = {B\"{u}ndelausgleichung f\"{u}r Multikamerasysteme}, year = {2011}, city = {Mainz}, proceeding = {Proceedings of the 31th DGPF Conference}, }

A temporal filter approach for real-time detection and reconstruction of curbs and road surfaces from 3D point clouds is presented. Instead of local thresholding, as used in many other approaches, a 3D curb model is extracted from the point cloud. The 3D points are classified to different parts of the model (i.e. road and sidewalk) using a temporally integrated Conditional Random Field (CRF). The parameters of curb and road surface are then estimated from the respectively assigned points, providing a temporal connection via a Kalman filter. In this contribution, we employ dense stereo vision for data acquisition. Other sensors capturing point cloud data, e.g. lidar, would also be suitable. The system was tested on real-world scenarios, showing the advantages over a temporally unfiltered version, due to robustness, accuracy and computation time. Further, the lateral accuracy of the system is evaluated. The experiments show the system to yield highly accurate results, for curved and straight-line curbs, up to distances of 20 meters from the camera.

@INPROCEEDINGS{siegemund*11:curb, author = {Jan Siegemund and Uwe Franke and Wolfgang F\"{o}rstner}, title = {A Temporal Filter Approach for Detection and Reconstruction of Curbs and Road Surfaces based on Conditional Random Fields}, booktitle = {IEEE Intelligent Vehicles Symposium (IV)}, year = {2011}, pages = {637-642}, month = {June}, publisher = {IEEE Computer Society} }

@ARTICLE{yang*11:fusion, author = {Yang, Michael Ying and Cao, Yanpeng and McDonald, John}, title = {{Fusion of camera images and laser scans for wide baseline 3D scene alignment in urban environments}}, journal = {ISPRS Journal of Photogrammetry and Remote Sensing}, year = {2011}, volume = {}, pages = {accepted}, }

@INPROCEEDINGS{yang*11:crf-classification, author = {Yang, Michael Ying and Förstner, Wolfgang}, title = {Regionwise Classification of Building Facade Images}, booktitle = {Photogrammetric Image Analysis (PIA2011)}, year = {2011}, pages = {209 -- 220}, publisher = {Springer}, series = {LNCS 6952}, }

@INPROCEEDINGS{yang*11:hcrf-scene-classification, author = {Yang, Michael Ying and Förstner, Wolfgang}, title = {A Hierarchical Conditional Random Field Model for Labeling and Classifying Images of Man-made Scenes}, booktitle = {International Conference on Computer Vision, IEEE/ISPRS Workshop on Computer Vision for Remote Sensing of the Environment}, year = {2011}, pages = {accepted}, }

This article addresses the reliable tracking of oncoming traffic at urban intersections from a moving platform with a stereo vision system. Both motion and depth information is combined to estimate the pose and motion parameters of an oncoming vehicle, including the yaw rate, by means of Kalman filtering. Vehicle tracking at intersections is particularly chal- lenging since vehicles can turn quickly. A single filter approach cannot cover the dynamic range of a vehicle sufficiently. We propose a real-time multi-filter approach for vehicle tracking at intersections. A gauge consistency criteria as well as a robust outlier detection method allow for dealing with sudden accelerations and self-occlusions during turn maneuvers. The system is evaluated both on synthetic and real-world data.

@INPROCEEDINGS{barth*10:itsc, author = {A. Barth and U. Franke}, title = {Tracking Oncoming and Turning Vehicles at Intersections}, booktitle = {Intelligent Transportation Systems, IEEE Conference on}, year = {2010}, pages = {861-868}, address = {Madeira Island, Portugal} }

A multi-class traffic scene segmentation approach based on scene flow data is presented. Opposed to many other approaches using color or texture features, our approach is purely based on dense depth and 3D motion information. Using prior knowledge on tracked objects in the scene and the pixel-wise uncertainties of the scene flow data, each pixel is assigned to either a particular moving object class (tracked/unknown object), the ground surface, or static background. The global topological order of classes, such as objects are above ground, is locally integrated into a conditional random field by an ordering constraint. The proposed method yields very accurate segmentation results on challenging real world scenes, which we made publicly available for comparison.

@inproceedings{barth*10:probabilistic, author={Barth, A. and Siegemund, J. and Mei{\ss}ner, A. and Franke, U. and F\"{ o}rstner, W.}, title={{Probabilistic Multi-Class Scene Flow Segmentation for Traffic Scenes }}, booktitle={Pattern Recognition (Symposium of DAGM) '10}, editor = {Goesele, M. and Roth, S. and Kuijper, A. and Schiele, B. and Schindler, K.}, year={2010}, publisher = {Springer} }

@ARTICLE {ceamanos*10: a classifier, author={Ceamanos,X. and Waske,B. and Benediktsson,J.A. and Chanussot,J. and Fauvel,M. and Sveinsson, J.R.}, title={{A classifier ensemble based on fusion of support vector machines for classifying hyperspectral data}}, journal={International Journal of Image and Data Fusion}, number={4}, pages={293-307}, year={2010} }

@ARTICLE{dalla mura*10:extended, author={Dalla Mura,M. and Benediktsson,J.A. and Waske,B. and Bruzzone,L.}, title={{Extended profiles with morphological attribute filters for the analysis of hyperspectral data}}, journal={International Journal of Remote Sensing}, number={22}, pages={5975-5991}, year={2010} }

Morphological attribute profiles (APs) are defined as a generalization of the recently proposed morphological profiles (MPs). APs provide a multilevel characterization of an image created by the sequential application of morphological attribute filters that can be used to model different kinds of the structural information. According to the type of the attributes considered in the morphological attribute transformation, different parametric features can be modeled. The generation of APs, thanks to an efficient implementation, strongly reduces the computational load required for the computation of conventional MPs. Moreover, the characterization of the image with different attributes leads to a more complete description of the scene and to a more accurate modeling of the spatial information than with the use of conventional morphological filters based on a predefined structuring element. Here, the features extracted by the proposed operators were used for the classification of two very high resolution panchromatic images acquired by Quickbird on the city of Trento, Italy. The experimental analysis proved the usefulness of APs in modeling the spatial information present in the images. The classification maps obtained by considering different APs result in a better description of the scene (both in terms of thematic and geometric accuracy) than those obtained with an MP.

@ARTICLE{dalla mura*10:morphological, author={Dalla Mura,M. and Benediktsson,J.A. and Waske,B. and Bruzzone,L.}, title={{ Morphological Attribute Profiles for the Analysis of Very High Resolution Images }}, journal={IEEE Transactions on Geoscience and Remote Sensing}, volume={48}, number={10}, pages={3747-3762}, year={2010}, }

Estimation using homogeneous entities has to cope with obstacles such as singularities of covariance matrices and redundant parametrizations which do not allow an immediate definition of maximum likelihood estimation and lead to estimation problems with more parameters than necessary. The paper proposes a representation of the uncertainty of all types of geometric entities and estimation procedures for geometric entities and transformations which (1) only require the minimum number of parameters, (2) are free of singularities, (3) allow for a consistent update within an iterative procedure, (4) enable to exploit the simplicity of homogeneous coordinates to represent geometric constraints and (5) allow to handle geometric entities which are at innity or at least very far, avoiding the usage of concepts like the inverse depth. Such representations are already available for transformations such as rotations, motions (Rosenhahn 2002), homographies (Begelfor 2005), or the projective correlation with fundamental matrix (Bartoli 2004) all being elements of some Lie group. The uncertainty is represented in the tangent space of the manifold, namely the corresponding Lie algebra. However, to our knowledge no such representations are developed for the basic geometric entities such as points, lines and planes, as in addition to use the tangent space of the manifolds we need transformation of the entities such that they stay on their specific manifold during the estimation process. We develop the concept, discuss its usefulness for bundle adjustment and demonstrate (a) its superiority compared to more simple methods for vanishing point estimation, (b) its rigour when estimating 3D lines from 3D points and (c) its applicability for determining 3D lines from observed image line segments in a multi view setup.

@INPROCEEDINGS{foerstner10:minimal, author = {F\"{o}rstner, W.}, title = {{Minimal Representations for Uncertainty and Estimation in Projective Spaces}}, booktitle = {Proc. of Asian Conference on Computer Vision}, year = {2010} }

The paper presents a method for automatically and optimally determining the vanishing points of a single image, and in case the interior orientation is given, the rotation of an image with respect to the intrinsic coordinate system of a lego land scene. We perform rigorous testing and estimation in order to be as independent on control parameters as possible. This refers to (1) estimating vanishing points from line segments and the rotation matrix, (2) to testing during RANSAC and during boosting lines and (3) to classifying the line segments w. r. t. their vanishing point. Spherically normalized homogeneous coordinates are used for line segments and especially for vanishing points to allow for points at infinity. We propose a minimal representation for the uncertainty of homogeneous coordinates of 2D points and 2D lines and rotations to avoid the use of singular covariance matrices of observed line segments. This at the same time allows to estimate the parameters with a minimal representation. The vanishing point detection method is experimentally validated on a set of 292 images.

@INPROCEEDINGS{foerstner10:optimal, author = {F\"{o}rstner, W.}, title = {{Optimal Vanishing Point Detection and Rotation Estimation of Single Images of a Legolandscene}}, booktitle = {Int. Archives of Photogrammetry and Remote Sensing}, year = {2010}, organization = {ISPRS Symposium Comm. III, Paris}, location = {wf} }

We present a fast automatic reproducible method for 3d semantic segmentation of magnetic resonance images of the knee. We formulate a single global model that allows to jointly segment all classes. The model estimation was performed automatically without manual interaction and parameter tuning. The segmentation of a magnetic resonance image with 11 Mio voxels took approximately one minute. Our labeling results by far do not reach the performance of complex state of the art approaches designed to produce clinically relevant results. Our results could potentially be useful for rough visualization or initialization of computationally demanding methods. Our main contribution is to provide insights in possible strategies when employing global statistical models.

@INPROCEEDINGS{korc*10:nonparametric, author = {Kor{\v c}, F. and Schneider, D. and F{\" o}rstner, W.}, title = {{On Nonparametric Markov Random Field Estimation for Fast Automatic Segmentation of MRI Knee Data}}, booktitle = {Proceedings of the 4th Medical Image Analysis for the Clinic - A Grand Challenge workshop, MICCAI}, year = {2010} }

With an omnidirectional camera system, it is possible to take 360°views of the surrounding area at each camera position. These systems are used particularly in robotic applications, in autonomous navigation and supervision technology for ego­motion estimation. In addition to the visual capture of the environment itself, we can compute the parameters of orientation and position from image sequences, i.e. we get three parameters of position and three of orientation (yaw rate, pitch and roll angle) at each time of acquisition. The aim of the presented project is to investigate the quality of the spatial trajectory of a mobile survey vehicle from the recorded image sequences. In this paper, we explain the required photogrammetric background and show the advantages of omnidirectional camera systems for this task. We present the first results on our test set and discuss alternative applications for omnidirectional cameras.

@INPROCEEDINGS{muffert*10:estimation, author = {Muffert, M. and Siegemund, Jan and F\"{o}rstner, Wolfgang}, title = {The estimation of spatial positions by using an omnidirectional camera system}, booktitle = {2nd International Conference on Machine Control \& Guidance}, year = {2010}, pages = {95-104}, month = {March} }

We discuss the utility of dimensionality reduction algorithms to put data points in high dimensional spaces into correspondence by learning a transformation between assigned data points on a lower dimensional structure. We assume that similar high dimensional feature spaces are characterized by a similar underlying low dimensional structure. To enable the determination of an affine transformation between two data sets we make use of well-known dimensional reduction algorithms. We demonstrate this procedure for applications like classification and assignments between two given data sets and evaluate six well-known algorithms during several experiments with different objectives. We show that with these algorithms and our transformation approach high dimensional data sets can be related to each other. We also show that linear methods turn out to be more suitable for assignment tasks, whereas graph-based methods appear to be superior for classification tasks.

@INPROCEEDINGS{roscher*10, author = {Roscher, R. and Waske, B. and F\"{o}rstner, W."}, title = {{Kernel Discriminative Random Fields for Land Cover Classification}}, booktitle={6th IAPR Workshop on Pattern Recognition in Remote Sensing}, year = {2010}, city = {Istanbul, Turkey}, proceeding = {6th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS 2010)} }

This paper presents a generic framework for curb detection and reconstruction in the context of driver assistance systems. Based on a 3D point cloud, we estimate the parameters of a 3D curb model, incorporating also the curb adjacent surfaces, e.g. street and sidewalk. We apply an iterative two step approach. First, the measured 3D points, e.g., obtained from dense stereo vision, are assigned to the curb adjacent surfaces using loopy belief propagation on a Conditional Random Field. Based on this result, we reconstruct the surfaces and in particular the curb. Our system is not limited to straight-line curbs, i.e. it is able to deal with curbs of different curvature and varying height. The proposed algorithm runs in real-time on our demon- strator vehicle and is evaluated in urban real-world scenarios. It yields highly accurate results even for low curbs up to 20 m distance.

@INPROCEEDINGS{siegemund*10:curb, author = {Jan Siegemund and David Pfeiffer and Uwe Franke and Wolfgang F\"{o}rstner}, title = {Curb Reconstruction using Conditional Random Fields}, booktitle = {IEEE Intelligent Vehicles Symposium (IV)}, year = {2010}, pages = {203-210}, month = {June}, publisher = {IEEE Computer Society} }

In this paper we propose a novel approach to bundle adjustment for large-scale camera configurations. The method does not need to include the 3D points in the optimization as parameters. Additionally, we model the parameters of a camera only relative to a nearby camera to achieve a stable estimation of all cameras. This guarantees to yield a normal equation system with a numerical condition, which practically is independent of the number of images. Secondly, instead of using the classical perspective relation between object point, camera and image point, we use epipolar and trifocal constraints to implicitly establish the relations between the cameras via the object structure. This avoids the explicit reference to 3D points thereby handling points far from the camera in a numerically stable fashion. We demonstrate the resulting stability and high convergence rates using synthetic and real data.

@INPROCEEDINGS{steffen*10:relative, author = {Steffen, R. and Frahm, J.-M. and F{\" o}rstner, W.}, title = {{Relative Bundle Adjustment based on Trifocal Constraints}}, booktitle = {ECCV Workshop on Reconstruction and Modeling of Large-Scale 3D Virtual Environments}, year = {2010}, organization = {ECCV 2010 Crete, Greece} }

The accuracy of supervised land cover classifications depends on factors such as the chosen classification algorithm, adequate training data, the input data characteristics, and the selection of features. Hyperspectral imaging provides more detailed spectral and spatial information on the land cover than other remote sensing resources. Over the past ten years, traditional and formerly widely accepted statistical classification methods have been superseded by more recent machine learning algorithms, e.g., support vector machines (SVMs), or by multiple classifier systems (MCS). This can be explained by limitations of statistical approaches with regard to high-dimensional data, multimodal classes, and often limited availability of training data. In the presented study, MCSs based on SVM and random feature selection (RFS) are applied to explore the potential of a synergetic use of the two concepts. We investigated how the number of selected features and the size of the MCS influence classification accuracy using two hyperspectral data sets, from different environmental settings. In addition, experiments were conducted with a varying number of training samples. Accuracies are compared with regular SVM and random forests. Experimental results clearly demonstrate that the generation of an SVM-based classifier system with RFS significantly improves overall classification accuracy as well as producer’s and user’s accuracies. In addition, the ensemble strategy results in smoother, i.e., more realistic, classification maps than those from stand-alone SVM. Findings from the experiments were successfully

@ARTICLE{5439692, author={Waske, B. and van der Linden, S. and Benediktsson, J. A. and Rabe, A. and Hostert, P.}, journal={Geoscience and Remote Sensing, IEEE Transactions on}, title={Sensitivity of Support Vector Machines to Random Feature Selection in Classification of Hyperspectral Data}, year={2010}, month={july }, volume={48}, number={7}, pages={2880 -2889}, keywords={}, doi={10.1109/TGRS.2010.2041784}, ISSN={0196-2892},}

In this paper, we point out the role of sequences of samples for training an incremental learning method. We define characteristics of incremental learning methods to describe the influence of sample ordering on the performance of a learned model. We show the influence of sequence for two different types of incremental learning. One is aimed on learning structural models, the other on learning models to discriminate object classes. In both cases, we show the possibility to find good sequences before starting the training.

@INPROCEEDINGS{wenzel*10:role, author = {Susanne Wenzel and Lothar Hotz}, title = {The Role of Sequences for Incremental Learning}, booktitle = {ICAART 2010 - Proceedings of the International Conference on Agents and Artificial Intelligence}, year = {2010}, editor = {Joaquim Filipe and Ana L. N. Fred and Bernadette Sharp}, volume = {1}, pages = {434-439}, address = {Valencia, Spain}, month = {January 22-24}, publisher = {INSTICC Press}, isbn = {978-989-674-021-4}, timestamp = {2011.01.18} }

@inproceedings{yang*10:alignment, author = {Yang, Michael Ying and Cao, Yanpeng and F\"{o}rstner, Wolfgang and McDonald, John}, title = {Robust wide baseline scene alignment based on 3D viewpoint normalization}, booktitle = {International Conference on Advances in Visual Computing}, year = {2010}, pages = {654--665}, publisher = {Springer-Verlag}, }

@TECHREPORT{yang*10:plane, author = {Yang, M. Y., Foerstner, W.}, title = {{Plane Detection in Point Cloud Data}}, institution = {Department of Photogrammetry, University of Bonn}, year = {2010}, number = {TR-IGG-P-2010-01 }

Multi-class image classification has made significant advances in recent years through the combination of local and global features. This paper proposes a novel approach called hierarchical conditional random field (HCRF) that explicitly models region adjacency graph and region hierarchy graph structure of an image. This allows to set up a joint and hierarchical model of local and global discriminative methods that augments conditional random field to a multi-layer model. Region hierarchy graph is based on a multi-scale watershed segmentation.

@INPROCEEDINGS{yang*10:hierarchical, author = {Yang, Michael Ying and F\"{o}rstner, Wolfgang and Martin Drauschke}, title = {{Hierarchical Conditional Random Field for Multi-class Image Classification}}, booktitle = {International Conference on Computer Vision Theory and Applications (VISSAPP)}, year = {2010}, pages = {464-469} }

Abstract. The (Extended) Kalman filter has been established as a stan- dard method for object tracking. While a constraining motion model stabilizes the tracking results given noisy measurements, it limits the ability to follow an object in non-modeled maneuvers. In the context of a stereo-vision based vehicle tracking approach, we propose and compare three different strategies to automatically adapt the dynamics of the fil- ter to the dynamics of the object. These strategies include an IMM-based multi-filter setup, an extension of the motion model considering higher order terms, as well as the adaptive parametrization of the filter vari- ances using an independent maximum likelihood estimator. For evalua- tion, various recorded real world trajectories and simulated maneuvers, including skidding, are used. The experimental results show significant improvements in the simultaneous estimation of pose and motion.

@INPROCEEDINGS{barth*09:simultaneous, author={Bart, A. and Siegemund, J. and Franke, U. and F{\" o}rstner, W.}, title={{Simultaneous Estimation of Pose and Motion at Highly Dynamic Turn Maneuvers}}, booktitle={Pattern Recognition (Symposium of DAGM) '09}, editor = {Denzler, J. and Notni, G.}, year={2009}, publisher = {Springer} }

This paper reports on methods for the automatic detection and classification of leaf diseases based on high resolution multispectral images. Leaf diseases are economically important as they could cause a yield loss. Early and reliable detection of leaf diseases therefore is of utmost practical relevance - especially in the context of precision agriculture for localized treatment with fungicides. Our interest is the analysis of sugar beet due to their economical impact. Leaves of sugar beet may be infected by several diseases, such as rust (Uromyces betae), powdery mildew (Erysiphe betae) and other leaf spot diseases (Cercospora beticola and Ramularia beticola). In order to obtain best classification results we apply conditional random fields. In contrast to pixel based classifiers we are able to model the local context and contrary to object centred classifiers we simultaneously segment and classify the image. In a first investigation we analyse multispectral images of single leaves taken in a lab under well controlled illumination conditions. The photographed sugar beet leaves are healthy or either infected with the leaf spot pathogen Cercospora beticola or with the rust fungus Uromyces betae. We compare the classification methods pixelwise maximum posterior classification (MAP), objectwise MAP as soon as global MAP and global maximum posterior marginal classification using the spatial context within a conditional random field model.

@INPROCEEDINGS{bauer*09:investigation, author = {Bauer, Sabine Daniela and Kor{\v c}, Filip and F\"{o}rstner, Wolfgang}, title = {{Investigation into the classification of diseases of sugar beet leaves using multispectral images}}, booktitle={Precision Agriculture 2009}, year = {2009}, pages = {229-238}, city = {Bonn}, proceeding = {Precision Agriculture} }

@INPROCEEDINGS{dickscheid*09:evaluating, author = {Timo Dickscheid and Wolfgang F\"{o}rstner}, title = {{Evaluating the Suitability of Feature Detectors for Automatic Image Orientation Systems}}, editor = {Mario Fritz and Bernt Schiele and Justus H. Piater}, booktitle = {Computer Vision Systems}, publisher = {Springer}, location = {Heidelberg}, series = {Lecture Notes in Computer Science}, volume = {5815}, year = {2009}, isbn = {978-3-642-04666-7}, pages = {305--314} }

@INPROCEEDINGS{drauschke09:irregular, author = {Drauschke, Martin}, title = {{An Irregular Pyramid for Multi-scale Analysis of Objects and their Parts}}, booktitle={7th IAPR-TC-15 Workshop on Graph-based Representations in Pattern Recognition}, year = {2009}, pages = {293-303}, city = {Bonn} }

@INPROCEEDINGS{drauschke*09:multidodging:, author = {Drauschke, Martin and F\"{o}rstner, Wolfgang and Brunn, Ansgar}, title = {Multidodging: Ein effizienter Algorithmus zur automatischen Verbesserung von digitalisierten Luftbildern}, booktitle={Publikationen der DGPF, Band 18: Zukunft mit Tradition}, year = {2009}, pages = {61-68}, city = {Bonn}, }

@inproceedings{drauschke*09:improving, author = {Drauschke, Martin and Roscher, Ribana and L\"{a}be, Thomas and F\"{o}rstner, Wolfgang}, title = {Improving Image Segmentation using Multiple View Analysis}, year = {2009}, pages = {211-216}, city = {Paris}, booktitle= {Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms and Evaluatin (CMRT09)} }

Photogrammetry has significantly been influenced by its two neigbouring fields, namely Computer Vision and Remote Sensing. Today, Photogrammetry has been become a part of Remote Sensing. The paper reflects its growing relations with Computer Vision, based on a more than 25 years experience of the author with the fascinating field between cognitive, natural and engineering science, which stimulated his own research and transferred him into a wanderer between two worlds.

@INPROCEEDINGS{foerstner09:computer, author = {F\"{o}rstner, Wolfgang}, title = {{Computer Vision and Remote Sensing - Lessons Learned}}, booktitle={Photogrammetric Week 2009}, year = {2009}, pages = {241-249}, city = {Stuttgart}, }

@inproceedings{foerstner09:mustererkennung, author = {F\"{o}rstner, Wolfgang}, title = {Mustererkennung in der Fernerkundung}, year = {2009}, pages = {129-136}, city = {Bonn}, booktitle= {Publikationen der DGPF, Band 18: Zukunft mit Tradition} }

@inproceedings{foerstner*09:completeness, author={W. F\"{o}rstner and T. Dickscheid and F. Schindler}, title={{On the Completeness of Coding with Image Features}}, booktitle={20th British Machine Vision Conference}, address={London, UK}, year={2009} }

@inproceedings{foerstner*09:detecting, author={W. F\"{o}rstner and T. Dickscheid and F. Schindler}, title={{Detecting Interpretable and Accurate Scale-Invariant Keypoints}}, booktitle={12th IEEE International Conference on Computer Vision (ICCV'09)}, address={Kyoto, Japan}, year={2009}, pages={2256-2263} }

@ARTICLE{meidow*09:reasoning, author = {Meidow, Jochen and Beder, Christian and F\"{o}rstner, Wolfgang}, title = {Reasoning with uncertain points, straight lines, and straight line segments in 2D}, journal={ISPRS Journal of Photogrammetry and Remote Sensing}, volume={64}, number={2}, year = {2009}, pages = {125-139}, city = {Bonn} }

Well known estimation techniques in computational geometry usually deal only with single geometric entities as unknown parameters and do not account for constrained observations within the estimation. The estimation model proposed in this paper is much more general, as it can handle multiple homogeneous vectors as well as multiple constraints. Furthermore, it allows the consistent handling of arbitrary covariance matrices for the observed and the estimated entities. The major novelty is the proper handling of singular observation covariance matrices made possible by additional constraints within the estimation. These properties are of special interest for instance in the calculus of algebraic projective geometry, where singular covariance matrices arise naturally from the non-minimal parameterizations of the entities. The validity of the proposed adjustment model will be demonstrated by the estimation of a fundamental matrix from synthetic data and compared to heteroscedastic regression [?], which is considered as state-ofthe- art estimator for this task. As the latter is unable to simultaneously estimate multiple entities, we will also demonstrate the usefulness and the feasibility of our approach by the constrained estimation of three vanishing points from observed uncertain image line segments.

@inproceedings{meidow*09:optimal, author={Meidow, J. and F{\" o}rstner, W. and Beder, C.}, title={{Optimal Parameter Estimation with Homogeneous Entities and Arbitrary Constraints}}, booktitle={Pattern Recognition (Symposium of DAGM) '09}, editor = {Denzler, J. and Notni, G.}, year={2009}, publisher = {Springer}, series = {LNCS} }

@article{schmittwilken*09:integration, author = {Schmittwilken, Jörg and Yang, Michael Ying and Förstner, Wolfgang and Plümer, Lutz}, title = {Integration of conditional random fields and attribute grammars for range data interpretation of man-made objects}, journal = {Annals of GIS}, volume = {15}, number = {2}, year = {2009}, pages = {117--126}, }

Hyperspectral imaging provides detailed spectral and spatial information from the land cover that enables a precise differentiation between various surface materials. On the other hand, the performance of traditional and widely used statistical classification methods is often limited in this context. Thus, alternative methods are required. In the study presented here, the performance of two machine learning techniques – Support Vector Machines (SVM) and Random Forests (RF) – are investigated and the classification results are compared to well known methods (i.e., maximum likelihood classifier and spectral angle mapper). The classifiers are applied to an AVIRIS data set that was acquired near the volcano Hekla in Iceland. The results clearly show the advantages of the two proposed classifier algorithms in terms of accuracy. They significantly outperform the other methods and achieve overall accuracies of approximately 90%. Although SVM and RF show some diversity in the classification results, the global performance of the two classifiers is very similar. Thus, both methods can be considered attractive for the classification of hyperspectral data.

@ARTICLE{waske*09:mapping, author = {Waske, Bj\"{o}rn and Benediktsson, Jon Atli and Árnason, Kolbein and Sveinsson, Johannes R.}, title = {Mapping of hyperspectral AVIRIS data using machine learning algorithms}, journal={Canadian Journal of Remote Sensing}, volume={35}, year = {2009}, number={S1}, pages = {106-116} }

SAR data are almost independent from weather conditions, and thus are well suited for mapping of seasonally changing variables such as land cover. In regard to recent and upcoming missions, multitemporal and multi-frequency approaches become even more attractive. In the present study, classifier ensembles (i.e., boosted decision tree and random forests) are applied to multi-temporal C-band SAR data, from different study sites and years. A detailed accuracy assessment shows that classifier ensembles, in particularly random forests, outperform standard approaches like a single decision tree and a conventional maximum likelihood classifier by more than 10% independently from the site and year. They reach up to almost 84% of overall accuracy in rural areas with large plots. Visual interpretation confirms the statistical accuracy assessment and reveals that also typical random noise is considerably reduced. In addition the results demonstrate that random forests are less sensitive to the number of training samples and perform well even with only a small number. Random forests are computationally highly efficient and are hence considered very well suited for land cover classifications of future multifrequency and multitemporal stacks of SAR imagery.

@article{Waske2009450, title = "Classifier ensembles for land cover mapping using multitemporal SAR imagery", journal = "ISPRS Journal of Photogrammetry and Remote Sensing", volume = "64", number = "5", pages = "450 - 457", year = "2009", note = "Theme Issue: Mapping with SAR: Techniques and Applications", issn = "0924-2716", doi = "DOI: 10.1016/j.isprsjprs.2009.01.003", url = "http://www.sciencedirect.com/science/article/B6VF4-4VP12G6-1/2/fc8ceb15fc4b40b2507ba5482ce204e8", author = "Björn Waske and Matthias Braun", keywords = "Decision tree", keywords = "Random forests", keywords = "Boosting", keywords = "Multitemporal SAR data", keywords = "Land cover classification" }

@ARTICLE{yang09:remote, author = {Yang, Ying}, title = {Remote sensing image registration via active contour model}, journal={International Journal of Electronics and Communications}, volume={65}, year = {2009}, pages = {227-234} }

We will present a novel incremental algorithm for the task of online least-squares estimation. Our approach aims at combining the accuracy of least-squares estimation and the fast computation of recursive estimation techniques like the Kalman filter. Analyzing the structure of least-squares estimation we devise a novel incremental algorithm, which is able to introduce new unknown parameters and observations into an estimation simultaneously and is equivalent to the optimal overall estimation in case of linear models. It constitutes a direct generalization of the well-known Kalman filter allowing to augment the state vector inside the update step. In contrast to classical recursive estimation techniques no artificial initial covariance for the new unknown parameters is required here. We will show, how this new algorithm allows more flexible parameter estimation schemes especially in the case of scene and motion reconstruction from image sequences. Since optimality is not guaranteed in the non-linear case we will also compare our incremental estimation scheme to the optimal bundle adjustment on a real image sequence. It will be shown that competitive results are achievable using the proposed technique.

@inproceedings{beder*08:incremental, author = {Beder, Christian and Steffen, Richard}, title = {{Incremental estimation without specifying a-priori covariance matrices for the novel parameters}}, year = {2008}, booktitle = {VLMP Workshop on CVPR} }

In classical photogrammetry, point observations are manually determined by an operator for performing the bundle adjustment of a sequence of images. In such cases, a comparison of different estimates is usually carried out with respect to the estimated 3D object points. Today, a broad range of automatic methods are available for extracting and matching point features across images, even in the case of widely separated views and under strong deformations. This allows for fully automatic solutions to the relative orientation problem, and even to the bundle triangulation in case that manually measured control points are available. However, such systems often contain random subprocedures like RANSAC for eliminating wrong correspondences, yielding different 3D points but hopefully similar orientation parameters. This causes two problems for the evaluation: First, the randomness of the algorithm has an influence on its stability, and second, we are constrained to compare the orientation parameters instead of the 3D points. We propose a method for benchmarking automatic bundle adjustments which takes these constraints into account and uses the orientation parameters directly. Given sets of corresponding orientation parameters, we require our benchmark test to address their consistency of the form deviation and the internal precision and their precision level related to the precision of a reference data set. Besides comparing different bundle adjustment methods, the approach may be used to safely evaluate effects of feature operators, matching strategies, control parameters and other design decisions for a particular method. The goal of this paper is to derive appropriate measures to cover these aspects, describe a coherent benchmarking scheme and show the feasibility of the approach using real data.

@INPROCEEDINGS{dickscheid*08:benchmarking, author = {Dickscheid, Timo and L{\"a}be, Thomas and F{\" o}rstner, Wolfgang}, title = {Benchmarking Automatic Bundle Adjustment Results}, booktitle = {To appear in: 21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, address = {Beijing, China}, year = {2008}, }

@INPROCEEDINGS{drauschke*08:comparison, author = {Drauschke, Martin and F{\" o}rstner, Wolfgang}, title = {Comparison of Adaboost and ADTboost for Feature Subset Selection}, booktitle = {PRIS 2008}, address = {Barcelona, Spain}, year = {2008}, }

The paper addresses the problem of feature selection during classification of image regions within the context of interpreting images showing highly structured objects such as buildings. We present a feature selection scheme that is connected with the classification framework Adaboost, cf. (Schapire and Singer, 1999). We constricted our weak learners on threshold classification on a single feature. Our experiments showed that the classification with Adaboost is based on relatively small subsets of features. Thus, we are able to find sets of appropriate features. We present our results on manually annotated and automatically segmented regions from facade images of the eTRIMS data base, where our focus were the object classes facade, roof, windows and window panes.

@INPROCEEDINGS{drauschke*08:selecting, author = {Drauschke, Martin and F{\" o}rstner, Wolfgang}, title = {Selecting appropriate features for detecting buildings and building parts}, booktitle = {To appear in: 21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, address = {Beijing, China}, year = {2008}, }

We investigate Discriminative Random Fields (DRF) which provide a principled approach for combining local discriminative classifiers that allow the use of arbitrary overlapping features, with adaptive data-dependent smoothing over the label field. We discuss the differences between a traditional Markov Random Field (MRF) formulation and the DRF model, and compare the performance of the two models and an independent sitewise classifier. Further, we present results suggesting the potential for performance enhancement by improving state of the art parameter learning methods. Eventually, we demonstrate the application feasibility on both synthetic and natural images.

@INPROCEEDINGS{korc*08:interpreting, author = {Kor{\v c}, Filip and F{\" o}rstner, Wolfgang}, title = {Interpreting Terrestrial Images of Urban Scenes Using Discriminative Random Fields}, booktitle = {Proc. of the 21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, year = {2008} }

We investigate maximum likelihood parameter learning in Conditional Random Fields (CRF) and present an empirical study of pseudo-likelihood (PL) based approximations of the parameter likelihood gradient. We show that these parameter learning methods can be improved and evaluate the resulting performance employing different inference techniques. We show that the approximation based on penalized pseudo-likelihood (PPL) in combination with the Maximum A Posteriori (MAP) inference yields results comparable to other state of the art approaches, while providing low complexity and advantages to formulating parameter learning as a convex optimization problem. Eventually, we demonstrate applicability on the task of detecting man-made structures in natural images.

@INPROCEEDINGS{korc*08:approximate, author = {Kor{\v c}, Filip and F{\" o}rstner, Wolfgang}, title = {Approximate Parameter Learning in {C}onditional {R}andom {F}ields: An Empirical Investigation}, booktitle = {Pattern Recognition}, year = {2008}, editor = {G. Rigoll}, number = {5096}, series = {LNCS}, pages = {11-20}, publisher = {Springer} }

We present a solution to the following discrete optimization problem. Given a set of independent, possibly overlapping image regions and a non-negative likeliness of the individual regions, we select a non-overlapping subset that is optimal with respect to the following requirements: First, every region is either part of the solution or has an overlap with it. Second, the degree of overlap of the solution with the rest of the regions is maximized together with the likeliness of the solution. Third, the likeliness of the individual regions influences the overall solution proportionally to the degree of overlap with neighboring regions. We represent the problem as a graph and solve the task by reduction to a constrained binary integer programming problem. The problem involves minimizing a linear objective function subject to linear inequality constraints. Both the objective function and the constraints exploit the structure of the graph. We illustrate the validity and the relevance of the proposed formulation by applying the method to the problem of facade window extraction. We generalize our formulation to the case where a set of hypotheses is given together with a binary similarity relation and similarity measure. Our formulation then exploits combination of degree and structure of hypothesis similarity and likeliness of individual hypotheses. In this case, we present a solution with non-similar hypotheses which can be viewed as a non-redundant representation.

@INPROCEEDINGS{korc*08:optimal, author = {Kor{\v c}, Filip and F{\" o}rstner, Wolfgang}, title = {Finding Optimal Non-Overlapping Subset of Extracted Image Objects}, booktitle = {Proc. of the 12th International Workshop on Combinatorial Image Analysis (IWCIA)}, year = {2008}, }

This paper presents an empirical investigation into the quality of automatic relative orientation procedures. The results of an in-house developed automatic orientation software called aurelo (Laebe and Foerstner, 2006) are evaluated. For this evaluation a recently proposed consistency measure for two sets of orientation parameters (Dickscheid et. al., 2008) and the ratio of two covariances matrices is used. Thus we evaluate the consistency of bundle block adjustments and the precision level achievable. We use different sets of orientation results related to the same set of images but computed under differing conditions. As reference datasets results on a much higher image resolution and ground truth data from artificial images rendered with computer graphics software are used. Six different effects are analysed: varying results due to random procedures in aurelo, computations on different image pyramid levels and with or without points with only two or three observations, the effect of replacing the used SIFT operator with an approximation of SIFT features, called SURF, repetitive patterns in the scene and remaining non-linear distortions. These experiments show under which conditions the bundle adjustment results reflect the true errors and thus give valuable hints for the use of automatic relative orientation procedures and possible improvements of the software.

@INPROCEEDINGS{laebe*08:quality, author = {L{\"a}be, Thomas and Dickscheid, Timo and F{\" o}rstner, Wolfgang}, title = {On the Quality of Automatic Relative Orientation Procedures}, booktitle = {To appear in: 21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, address = {Beijing, China}, year = {2008}, }

This paper addresses the challenge of a real-time capable vision system in the task of trajectory and surface reconstruction by aerial image sequences. The goal is to present the design, methods and strategies of a real-time capable vision system solving the mapping task for secure navigation of small UAVs with a single camera. This includes the estimation process, map representation, initialization processes, loop closing detection and exploration strategies. The estimation process is based on the Kalman-Filter and a landmark based map representation. We introduce a new initialization method for new observed landmarks. We will show that the initialization process and the exploration strategy has a significant effect on the accuracy of the estimated camera trajectory and of the map.

@INPROCEEDINGS{steffen*08:visual, author = {Steffen, Richard and F\"{o}rstner, Wolfgang}, title = {On Visual Real Time Mapping for Unmanned Aerial Vehicles}, year = "{2008}, pages = {57-62 Part B3a}, city = {Bonn}, booktitle= {21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)} }

A strategy for the joint classification of multiple segmentation levels from multisensor imagery is introduced by using synthetic aperture radar and optical data. At first, the two data sets are separately segmented, creating independent aggregation levels at different scales. Each individual level from the two sensors is then preclassified by a support vector machine (SVM). The original outputs of each SVM, i.e., images showing the distances of the pixels to the hyperplane fitted by the SVM, are used in a decision fusion to determine the final classes. The fusion strategy is based on the application of an additional classifier, which is applied on the preclassification results. Both a second SVM and random forests (RF) were tested for the decision fusion. The results are compared with SVM and RF applied to the full data set without preclassification. Both the integration of multilevel information and the use of multisensor imagery increase the overall accuracy. It is shown that the classification of multilevel-multisource data sets with SVM and RF is feasible and does not require a definition of ideal aggregation levels. The proposed decision fusion approach that applies RF to the preclassification outperforms all other approaches

@ARTICLE{4481234, title={Classifying Multilevel Imagery From SAR and Optical Sensors by Decision Fusion}, author={Waske, B. and van der Linden, S.}, journal={Geoscience and Remote Sensing, IEEE Transactions on}, year={2008}, month={May }, volume={46}, number={5}, pages={1457-1466}, doi={10.1109/TGRS.2008.916089}, ISSN={0196-2892}, }

We present a method for detecting repeated structures, which is applied on facade images for describing the regularity of their windows. Our approach finds and explicitly represents repetitive structures and thus gives initial representation of facades. No explicit notion of a window is used; thus, the method also appears to be able to identify other manmade structures, e.g., paths with regular tiles. A method for detection of dominant symmetries is adapted for detection of multiply repeated structures. A compact description of the repetitions is derived from the detected translations in the image by a heuristic search method and the criterion of the minimum description length.

@ARTICLE{wenzel*08:detection, author = {Wenzel, S. and Drauschke, M. and F{\" o}rstner, W.}, title = {Detection of repeated structures in facade images}, journal = {Pattern Recognition and Image Analysis}, year = {2008}, volume = {18}, pages = {406--411}, number = {3}, month = {September}, doi = {10.1134/S1054661808030073} }

We propose an incremental learning scheme for learning a class hierarchy for objects typically occurring multiple in images. Given one example of an object that appears several times in the image, e.g. is part of a repetitive structure, we propose a method for identifying prototypes using an unsupervised clustering procedure. These prototypes are used for building a hierarchical appearance based model of the envisaged class in a supervised manner. For classification of new instances detected in new images we use linear subspace methods that combine discriminative and reconstructive properties. The used methods are chosen to be capable for an incremental update. We test our approach on facade images with repetitive windows and balconies. We use the learned object models to find new instances in other images, e. g. the neighbouring facade and update already learned models with the new instances.

@INPROCEEDINGS{wenzel*08:semi-supervised, author = {Wenzel, Susanne and F{\" o}rstner, Wolfgang}, title = {Semi-supervised incremental learning of hierarchical appearance models}, booktitle = {21st Congress of the International Society for Photogrammetry and Remote Sensing (ISPRS)}, address = {Beijing, China}, year = {2008}, }

@INPROCEEDINGS{drauschke*07:automatic, author = {Drauschke, Martin and Brunn, Ansgar and Kulschewski, Kai and F{\"o}rstner, Wolfgang}, title = {Automatic Dodging of Aerial Images}, booktitle = {Publikationen der DGPF: Von der Medizintechnik bis zur Planetenforschung - Photogrammetrie und Fernerkundung für das 21. Jahrhundert}, year = {2007}, editor = {Seyfert, Eckhardt}, volume = {16}, pages = {173-180}, address = {Muttenz, Basel}, month = {June}, publisher = {DGPF}, }

This work contributes to detection and tracking of walking or running humans in surveillance video sequences. We propose a 2D model-based approach to the whole body tracking in a video sequence captured from a single camera view. An extended six-link biped human model is employed. We assume that a static camera observes the scene horizontally or obliquely. Persons can be seen from a continuum of views ranging from a lateral to a frontal one. We do not expect humans to be the only moving objects in the scene and to appear at the same scale at different image locations.

@INBOOK{KorcHlavacDetectTrackHumans2007, author = {Kor{\v c}, Filip and Hlav{\'a}{\v c}, V{\'a}clav}, year = {2007}, chapter = {Detection and Tracking of Humans in Single View Sequences Using 2D Articulated Model}, title = {Human Motion - Understanding, Modeling, Capture and Animation}, editor = {Rosenhahn, Bodo and Klette, Reinhard and Metaxas, Dimitris}, publisher = {Springer}, series = {Computational Imaging and Vision}, volume = {36}, edition = {1}, isbn = {978-1-4020-6692-4}, pages = {105--130}, keywords = {human detection in video, model-based human detection}, url = {ftp://cmp.felk.cvut.cz/pub/cmp/articles/hlavac/2007KorcHlavacDetectTrackHumans.pdf} }

@INPROCEEDINGS{saatkamp*07:generative, author = {Saatkamp, Jens and Schmittwilken, J{\"o}rg}, title = {Generative Models and Markov Chain Monte Carlo Techniques for Detection and Reconstruction of Stairs from Point Clouds}, booktitle = {Proceedings of the ISPRS Workshop on Updating Geo-spatial Databases with Imagery \& The 5th ISPRS Workshop on Dynamic and Multi-dimensional GIS}, year = {2007}, editor = {Jiang, Jie and Zhao, Renliang}, volume = {XXXVI}, number = {part 4/W54}, series = {The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences}, pages = {111-119}, address = {Urumqi, China}, month = {August}, organization = {ISPRS}, }

@ARTICLE{schmittwilken*07:semantic, author={Schmittwiken, J\"{o}rg and Saatkamp, Jens and F\"{o}rstner, Wolfgang and Kolbe, Thomas and Pl\"{u}mer, Lutz}, title={{A Semantic Model of Stairs in Building Collars}}, journal={Photogrammetrie, Fernerkundung, Geoinformation PFG}, year={2007}, pages={415-428}, }

Recursive estimation or Kalman filtering usually relies on explicit model functions, that directly and explicitly describe the effect of the parameters on the observations. However, many problems in computer vision, including all those resulting in homogeneous equation systems, are easier described using implicit constraints between the observations and the parameters. By implicit we mean, that the constraints are given by equations, that are not easily solvable for the observation vector. We present a framework, that allows to incorporate such implicit constraints as measurement equations into a Kalman filter. The algorithm may be used as a black-box, simplifying the process of specifying suitable measurement equations for many problems. As a byproduct, the possibility of specifying model equations non-explicitly, some non-linearities may be avoided and better results can be achieved for certain problems.

@INPROCEEDINGS{steffen*07:recursive, author = {Richard Steffen and Christian Beder}, title = {Recursive Estimation with Implicit Constraints}, booktitle = {Proceedings of the DAGM 2007}, editor = {F.A. Hamprecht and C. Schn\"orr and B. J\"ahne}, number = {4713}, year = {2007}, series = {LNCS}, publisher = {Springer}, pages = {194-203}, }

Large scale imagery will be increasingly available due to the low cost of video cameras and unmanned aerial vehicles. Their use is broad: the documentation of traffic accidents, the effects of thunderstorms onto agricultural farms, the 3Dstructure of industrial plants or the monitoring of archeological excavation. The value of imagery depends on the availability of (1) information about the place and date during data capture, (2) of information about the 3D-structure of the object and (3) of information about the class or identity of the objects in the scene. Geocoding, problem (1), usually relies the availability of GPS-information, which however limits the use of imagery to outdoor applications. The paper discusses methods for geocoding and geometrical evaluation of such imagery and especially adresses the question in how far the methods can do without GPS.

@ARTICLE{foerstner*07:online, author = {Wolfgang F{\"o}rstner and Richard Steffen}, title = {Online geocoding and evaluation of large scale imagery without GPS}, journal = {Photogrammetric Week, Heidelberg}, year = {2007}, volume = {Wichmann Verlag}, editor = {D. Fritsch}, }

The classification of multisensor data sets, consisting of multitemporal synthetic aperture radar data and optical imagery, is addressed. The concept is based on the decision fusion of different outputs. Each data source is treated separately and classified by a support vector machine (SVM). Instead of fusing the final classification outputs (i.e., land cover classes), the original outputs of each SVM discriminant function are used in the subsequent fusion process. This fusion is performed by another SVM, which is trained on the a priori outputs. In addition, two voting schemes are applied to create the final classification results. The results are compared with well-known parametric and nonparametric classifier methods, i.e., decision trees, the maximum-likelihood classifier, and classifier ensembles. The proposed SVM-based fusion approach outperforms all other approaches and significantly improves the results of a single SVM, which is trained on the whole multisensor data set.

@ARTICLE{4358858, title={Fusion of Support Vector Machines for Classification of Multisensor Data}, author={Waske, B. and Benediktsson, J.A.}, journal={Geoscience and Remote Sensing, IEEE Transactions on}, year={2007}, month={Dec. }, volume={45}, number={12}, pages={3858-3866}, doi={10.1109/TGRS.2007.898446}, ISSN={0196-2892}, }

We present a method for detecting repeated structures, which is applied on facade images for describing the regularity of their windows. Our approach finds and explicitly represents repetitive structures and thus gives initial representation of facades. No explicit notion of a window is used, thus the method also appears to be able to identify other man made structures, e.g. paths with regular tiles. A method for detection of dominant symmetries is adapted for detection of multiply repeated structures. A compact description of the repetitions is derived from the detected translations in the image by a heuristic search method and the criterion of the minimum description length.

@INPROCEEDINGS{wenzel*07a:detection, author = {Wenzel, S. and Drauschke, M. and F{\" o}rstner, W.}, title = {{Detection of repeated structures in facade images}}, booktitle = {Proceedings of the OGRW-7-2007, 7th Open German/Russian Workshop on Pattern Recognition and Image Understanding. August 20-23, 2007. Ettlingen, Germany}, year = {2007} }

We present a method for detecting repeated structures, which is applied on facade images for describing the regularity of their windows. Our approach finds and explicitly represents repetitive structures and thus gives initial representation of facades. No explicit notion of a window is used, thus the method also appears to be able to identify other man made structures, e.g. paths with regular tiles. A method for detection of dominant symmetries is adapted for detection of multiple repeated structures. A compact description of repetitions is derived from translations detected in an image by a heuristic search method and the model selection criterion of the minimum description length.

@ARTICLE{wenzel*07b:detection, author = {Wenzel, S. and Drauschke, M. and F{\" o}rstner, W.}, title = {{Detection and Description of Repeated Structures in Rectified Facade Images}}, journal = {Photogrammetrie, Fernerkundung, Geoinformation (PFG)}, year = {2007}, volume = {7}, pages = {481-490} }

Regelmäßige Strukturen und Symmetrien kennzeichnen viele Gebäudefassaden oder Objekte im Umfeld von Gebäuden. Für die automatisierte Bildinterpretation weisen diese Strukturen auf künstliche Objekte hin, führen aber auch zu Schwierigkeiten bei klassischen Bildzuordnungsverfahren. Die Suche und Gruppierung zusammengehöriger Merkmale kann daher sowohl zur Identifikation künstlicher Objekte als auch zur Verbesserung von Zuordnungsverfahren dienen. Für die Analyse von entzerrten Fassadenaufnahmen haben wir das Verfahren von [LOY 2006] zur Detektion symmetrischer Bildstrukturen zu einem Verfahren zur Detektion verschiedener, sich wiederholender Bildstrukturen erweitert und aus den detektierten wiederholten Objekten eine minimale Beschreibung der Struktur der Fassadenelemente in Form von achsenparallelen Basiselementen abgeleitet.

@INPROCEEDINGS{wenzel*07:detektion, author = {Wenzel, Susanne and Drauschke, Martin and F{\"o}rstner, Wolfgang}, title = {Detektion wiederholter und symmetrischer Strukturen in Fassadenbildern}, booktitle = {Publikationen der DGPF: Von der Medizintechnik bis zur Planetenforschung - Photogrammetrie und Fernerkundung für das 21. Jahrhundert}, year = {2007}, editor = {Seyfert, Eckhardt}, volume = {16}, pages = {119-126}, address = {Muttenz, Basel}, month = {June}, publisher = {DGPF}, }

Efficient direct solutions for the determination of a cylinder from points are presented. The solutions range from the well known direct solution of a quadric to the minimal solution of a cylinder with five points. In contrast to the approach of G. Roth and M. D. Levine (1990), who used polynomial bases for representing the geometric entities, we use algebraic constraints on the quadric representing the cylinder. The solutions for six to eight points directly determine all the cylinder parameters in one step: (1) The eight-point-solution, similar to the estimation of the fundamental matrix, requires to solve for the roots of a 3rd-order-polynomial. (2) The seven-point-solution, similar to the sixpoint- solution for the relative orientation by J. Philip (1996), yields a linear equation system. (3) The six-point-solution, similar to the fivepoint- solution for the relative orientation by D. Nister (2003), yields a ten-by-ten eigenvalue problem. The new minimal five-point-solution first determines the direction and then the position and the radius of the cylinder. The search for the zeros of the resulting 6th order polynomials is e ciently realized using 2D-Bernstein polynomials. Also direct solutions for the special cases with the axes of the cylinder parallel to a coordinate plane or axis are given. The method is used to find cylinders in range data of an industrial site.

@INPROCEEDINGS{beder*06:direct, author = {Christian Beder and Wolfgang F{\"o}rstner}, title = {Direct Solutions for Computing Cylinders from Minimal Sets of 3D Points}, booktitle = {Proceedings of the European Conference on Computer Vision}, year = {2006}, editor = {A. Leonardis and H. Bischof and A. Pinz}, number = {3951}, series = {LNCS}, pages = {135-146}, publisher = {Springer}, }

Die automatische Extraktion von Zylindern aus 3D-Punktwolken ist von zentraler Bedeutung bei der Auswertung von Laserscannerdaten insbesondere bei Industrieanlagen. Das robuste Schätzverfahren RANSAC benötigt direkte Lösungen aus so wenig Datenpunkten wie möglich, um effizient zu arbeiten. Wir werden die algebraischen Bedingungen, die quadratische Formen erfüllen müssen, um einen Zylinder darzustellen, analysieren und verschiedene Verfahren für die Lösung dieses Problems vorstellen. Insbesondere werden wir eine minimale Lösung mit nur fünf 3D Punkten präsentieren. Anders als andere Ansätze benötigen wir keine Oberflächennormalen, deren Bestimmung im Allgemeinen schwierig ist.

@INPROCEEDINGS{beder*06:direkte, author = {Christian Beder and Wolfgang F{\"o}rstner}, title = {{D}irekte {B}estimmung von {Z}ylindern aus 3{D}-{P}unkten ohne {N}utzung von {O}berfl{\"a}chennormalen}, booktitle = {{P}hotogrammetrie - {L}aserscanning - {O}ptische 3{D}-{M}esstechnik}, year = {2006}, pages = {206-213}, publisher = {Herbert Wichmann Verlag}, editor = {Thomas Luhmann and Christina M{\"u}ller}, }

Algorithms for metric 3d reconstruction of scenes from calibrated image sequences always require an initialization phase for fixing the scale of the reconstruction. Usually this is done by selecting two frames from the sequence and fixing the length of their base-line. In this paper a quality measure, that is based on the uncertainty of the reconstructed scene points, for the selection of such a stable image pair is proposed. Based on this quality measure a fully automatic initialization phase for simultaneous localization and mapping algorithms is derived. The proposed algorithm runs in real-time and some results for synthetic as well as real image sequences are shown.

@INPROCEEDINGS{beder*06:determining, author = {Christian Beder and Richard Steffen}, title = {Determining an initial image pair for fixing the scale of a 3d reconstruction from an image sequence}, booktitle = {Pattern Recognition}, editor = {K. Franke and K.-R. M{\"u}ller and B. Nickolay and R. Schäfer}, number = {4174}, year = {2006}, series = {LNCS}, publisher = {Springer}, pages = {657-666}, }

Für die automatische Erfassung von Gebäuden aus Luftbildern ist es nützlich, Bildstrukturen im Skalenraum, d. h. über mehrere Auflösungsstufen zu beobachten, um für die Objekterkennung hinderliche Details ausblenden zu können. Große Bedeutung messen wir dabei den homogenen Regionen sowie deren Nachbarschaften zu. Regionen betrachten wir als stabil, wenn sie über einen mehrere Skalenstufen invariant bleiben. Sie haben spezielle Eigenschaften: Beim Vergrössern der Skala verschmelzen benachbarte Regionen, wobei eine Region immer vollständig in der anderen aufgeht. Diese speziellen Eigenschaft erleichtert das Bestimmen der Nachbarschaften in einer vorgegeben Skala, denn der Regionennachbarschaftsgraph (RNG) muss nur einmal auf der untersten Ebene des Skalenraums berechnet werden. Die RNGs in den anderen Ebenen können leicht aus der untersten Ebene berechnet werden.

@INPROCEEDINGS{drauschke*06:stabilitaet, author = {Martin Drauschke and Hanns-Florian Schuster and Wolfgang F{\"o}rstner}, title = {Stabilit{\"a}t von Regionen im Skalenraum}, booktitle = {Publikationen der DGPF: Geoinformatik und Erdbeobachtung}, year = {2006}, editor = {Eckhardt Seyfert}, volume = {15}, pages = {29-36}, address = {Berlin}, month = {Septermber}, publisher = {DGPF}, abstract = {F{\"u}r die automatische Erfassung von Geb{\"a}uden aus Luftbildern ist es n{\"u}tzlich, Bildstrukturen im Skalenraum, d. h. {\"u}ber mehrere Aufl{\"o}sungsstufen zu beobachten, um für die Objekterkennung hinderliche Details ausblenden zu k{\"o}nnen. Große Bedeutung messen wir dabei den homogenen Regionen sowie deren Nachbarschaften zu. Regionen betrachten wir als stabil, wenn sie über einen mehrere Skalenstufen invariant bleiben. Sie haben spezielle Eigenschaften: Beim Vergr{\"o}ssern der Skala verschmelzen benachbarte Regionen, wobei eine Region immer vollst{\"a}ndig in der anderen aufgeht. Diese speziellen Eigenschaft erleichtert das Bestimmen der Nachbarschaften in einer vorgegeben Skala, denn der Regionennachbarschaftsgraph (RNG) muss nur einmal auf der untersten Ebene des Skalenraums berechnet werden. Die RNGs in den anderen Ebenen k{\"o}nnen leicht aus der untersten Ebene berechnet werden.}, }

Automatic scene interpretation of aerial images is a major purpose of photogrammetry. Therefore, we want to improve building detection by exploring the "life-time" of stable and relevant image features in scale space. We use watersheds for feature extraction to gain a topologically consistent map. We will show that characteristic features for building detection can be found in all considered scales, so that no optimal scale can be selected for building recognition. Nevertheless, many of these features "live" in a wide scale interval, so that a combination of a small number of scales can be used for automatic building detection.

@INPROCEEDINGS{drauschke*06:detectibility, author = {Martin Drauschke and Hanns-Florian Schuster and Wolfgang F{\"o}rstner}, title = {Detectibility of Buildings in Aerial Images over Scale Space}, booktitle = {Symposium of ISPRS Commission III: Photogrammetric Computer Vision}, year = {2006}, editor = {Wolfgang F{\"o}rstner and Richard Steffen}, volume = {XXXVI}, number = {Part 3}, pages = {7-12}, address = {Bonn}, month = {20-22 September 2006}, organization = {ISPRS}, publisher = {ISPRS}, abstract = {Automatic scene interpretation of aerial images is a major purpose of photogrammetry. Therefore, we want to improve building detection by exploring the "life-time" of stable and relevant image features in scale space. We use watersheds for feature extraction to gain a topologically consistent map. We will show that characteristic features for building detection can be found in all considered scales, so that no optimal scale can be selected for building recognition. Nevertheless, many of these features "live" in a wide scale interval, so that a combination of a small number of scales can be used for automatic building detection.}, keywords = {Building Detection, Scale Space, Feature Extraction, Stable Regions}, }

This paper presents a new full automatic approach for the relative orientation of several digital images taken with a calibrated camera. This approach uses new algorithms for feature extraction and relative orientation developed in the last few years. There is no need for special markers in the scene nor for approximate values of the orientation data. We use the point operator developed by D. G. Lowe (2004), which extracts points with scale- and rotation-invariant descriptors (SIFT-features). These descriptors allow a successful matching of image points even when dealing with highly convergent or rotated images. The approach consists of the following steps: After extracting image points on all images a matching between every image pair is calculated using the SIFT parameters only. No prior information about the pose of the images or the overlapping parts of the images is used. For every image pair a relative orientation is computed with the help of a RANSAC procedure. Here we use the new 5-point algorithm from D. Nister (2004). Out of this set of orientations approximate values for the orientation parameters and the object coordinates are calculated by computing the relative scales and transforming the models into a common coordinate system. Several tests are made in order to get a reliable input for the currently final step: a bundle block adjustment. The paper discusses the practical impacts of the used algorithms. Examples of different indoor- and outdoor-scenes including a data set of oblique images taken from a helicopter are presented and the results of the approach applied to these data sets are evaluated. These results show that the approach can be used for a wide range of scenes with different types of the image geometry and taken with different types of cameras including inexpensive consumer cameras. In particular we investigate in the robustness of the algorithms, e. g. in geometric tests on image triplets. Further developments like the use of image pyramids with a modified matching are discussed in the outlook. Literature: David G. Lowe, Distinctive image features from scale-invariant keypoints, International Journal of Computer Vision, 60, 2 (2004), pp. 91-110. D. Nistér, An efficient solution to the five-point relative pose problem, IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 26(6):756-770, June 2004.

@inproceedings{läbe*06:automatic, author = "Läbe, Thomas and Förstner, Wolfgang", title = "AUTOMATIC RELATIVE ORIENTATION OF IMAGES", year = "06", city = "Bonn", proceeding = "Proceedings of the 5th Turkish-German Joint Geodetic Days",}

In the proposed approach, the well-known enhanced Lee filter is modified to allow the integration of feature outlines-previously extracted from segmented optical images. The filter is applied to several ENVISAT ASAR images that cover urban, agricultural, and forest areas during different plant phenological stages. The performance of this segment-based speckle filter is compared to those of other filters using ratio images, visual interpretation, and statistical indexes. The approach reduces the loss of radiometry and spatial information. It performs comparable to more complex methods and outperforms common techniques

@ARTICLE{4156176, title={A Segment-Based Speckle Filter Using Multisensoral Remote Sensing Imagery}, author={Waske, B. and Braun, M. and Menz, G.}, journal={Geoscience and Remote Sensing Letters, IEEE}, year={2007}, month={April }, volume={4}, number={2}, pages={231-235}, doi={10.1109/LGRS.2006.888849}, ISSN={1545-598X},

The paper describes calibration and epipolar rectification for stereo with fish-eye optics. While stereo processing of classical cameras is state of the art for many applications, stereo with fish-eye cameras have been much less discussed in literature. This paper discusses the geometric calibration and the epipolar rectification as pre-requisite for stereo processing with fish-eyes. First, it surveys mathematical models to describe the projection. Then the paper presents a method of generating epipolar images which are suitable for stereo-processing with a field of view larger than 180° in vertical and horizontal viewing directions. One example with 3D-point measuring from real fish-eye images demonstrates the feasibility of the calibration and rectification procedure. *Keywords: *fish-eye camera calibration; fish-eye stereo; epipolar rectification

An information theoretic framework for grouping observations is proposed. The entropy change incurred by new observations is analyzed using the Kalman filter update equations. It is found, that the entropy variation is caused by a positive similarity term and a negative proximity term. Bounding the similarity term in the spirit of the minimum description length principle and the proximity term in the spirit of maximum entropy inference a robust and efficient grouping procedure is devised. Some of its properties are demonstrated for the exemplary task of edgel grouping.

@INPROCEEDINGS{beder05:agglomerative, author = {Christian Beder}, title = {Agglomerative Grouping of Observations by Bounding Entropy Variation}, booktitle = {Pattern Recognition}, year = {2005}, editor = {Kropatsch, Walter and Sablatnig, Robert and Hanbury, Allan}, number = {3663}, series = {LNCS}, pages = {101-108}, organization = {DAGM}, publisher = {Springer}, pdf = {http://www.ipb.uni-bonn.de/papers/2005/beder05.agglomerative.pdf}, }

Geometric reasoning in Computer Vision always is performed under uncertainty. The great potential of both, projective geometry and statistics, can be integrated easily for propagating uncertainty through reasoning chains, for making decisions on uncertain spatial relations and for optimally estimating geometric entities or transformations. This is achieved by (1) exploiting the potential of statistical estimation and testing theory and by (2) choosing a representation of projective entities and relations which supports this integration. The redundancy of the representation of geometric entities with homogeneous vectors and matrices requires a discussion on the equivalence of uncertain projective entities. The multi-linearity of the geometric relations leads to simple expressions also in the presence of uncertainty. The non-linearity of the geometric relations finally requires to analyze the degree of approximation as a function of the noise level and of the embedding of the vectors in projective spaces. The paper discusses a basic link of statistics and projective geometry, based on a carefully chosen representation, and collects the basic relations in 2D and 3D and for single view geometry.

@InCollection{foerstner04:uncertainty, author = {W. F{\"o}rstner}, title = {Uncertainty and Projective Geometry}, booktitle = {Handbook of Computational Geometry for Pattern Recognition, Computer Vision, Neurocomputing and Robotics}, OPTcrossref = {}, OPTkey = {}, OPTpages = {to appear}, publisher = {Springer}, year = {2004}, editor = {E. Bayro-Corrochano}, OPTvolume = {}, OPTnumber = {}, OPTseries = {}, OPTannote = {} }

Der Aufsatz präsentiert ein neues vollautomatisches Verfahren zur relativen Orientierung mehrerer digitaler Bilder kalibrierter Kameras. Es nutzt die in den letzten Jahren neu entwickelten Algorithmen im Bereich der Merkmalsextraktion und der Bildgeometrie und erfordert weder das Anbringen von künstlichen Zielmarken noch die Angabe von Näherungswerten. Es basiert auf automatisch extrahierten Punkten, die mit dem von D. Lowe (2004) vorgeschlagenen Verfahren zur Extraktion skaleninvarianter Bildmerkmale berechnet werden. Diese ermöglichen eine Punktzuordnung auch bei stark konvergenten Aufnahmen. Für die Bestimmung von Näherungswerten der abschließenden Bündelausgleichung wird bei der relativen Orientierung der Bildpaare das direkte Lösungsverfahren von D. Nister (2004) verwendet. Der Aufsatz diskutiert die praktischen Erfahrungen mit den verwendeten Algorithmen anhand von Beispieldatensätzen sowohl von Innenraum- als auch von Aussnaufnahmen.

@inproceedings{läbe*05:erfahrungen, author = "Läbe, Thomas and Förstner, Wolfgang", title = "Erfahrungen mit einem neuen vollautomatischen Verfahren zur Orientierung digitaler Bilder", year = "05", city = "Bonn", proceeding = "Proceedings of DGPF Conference",}

Automatic quality evaluation of photogrammetric building acquisitions is important to realize deficiencies of acquisition approaches, tocompare different acquisitions approaches and to check the keeping of contractual specifications. For the decision-makers a procedure will be suggested taking a few, good interpretable quality measures into account. Therefore, useful quality measures have to be identifiedby the formulation of criteria. These quantities can be derived from the comparison of a test data set and a reference data set capturing the same scene. The acquired topology is usually uncertain as for instance two adjacent buildings may be acquired as one building ortwo buildings. Thus a screening of the registered area is suggested to compute the quantities. The approach is independent of the used acquisition method. For the application of large data sets the corresponding data structures will be explained. In experimental tests thebuildings registered by two commercial acquisition systems will be compared by the quality measures determined in 2D and 3D.

@inproceedings{meidow*05:voxel-based, author = "Meidow, Jochen and Schuster, Hanns-Florian", title = "Voxel-based Quality Evaluation of Photogrammetic Buildingacquisitions", year = "05", city = "Bonn", proceeding = "Proceedings of the CMRT05-Workshop",}

This work addresses the two major drawbacks of current statistical uncertain geometric reasoning approaches. In the first part a framework is presented, that allows to represent uncertain line segments in 2D- and 3D-space and perform statistical test with these practically very important types of entities. The second part addresses the issue of performance of geometric reasoning. A data structure is introduced, that allows the efficient processing of large amounts of statistical tests involving geometric entities. The running times of this approach are finally evaluated experimentally.

@INPROCEEDINGS{beder04:fast, author = {Christian Beder}, title = {Fast Statistically Geometric Reasoning About Uncertain Line Segments in 2D- and 3D-Space}, booktitle = {Pattern Recognition}, year = {2004}, editor = {C.E.Rasmussen and H.H.Bülthoff and B.Schölkopf and M.A.Giese}, number = {3175}, series = {LNCS}, pages = {375-382}, organization = {DAGM}, publisher = {Springer}, pdf = {http://www.ipb.uni-bonn.de/papers/2004/beder04.fast.pdf}, }

The accurate reconstruction of the three-dimensional structure from multiple images is still a challenging problem, so that most current approaches are based on semi-automatic procedures. Therefore the introduction of accurate and reliable automation for this classical problem is one of the key goals of photogrammetric research. This work deals with the problem of matching points and lines across multiple views, in order to gain a highly accurate reconstruction of the depicted object in three-dimensional space. In order to achieve this goal, a novel framework is introduced, that draws a sharp boundary between feature extraction, feature matching based on geometric constraints and feature matching based on radiometric constraints. The isolation of this three parts allows direct control and therefore better understanding of the different kinds of influences on the results. Most image feature matching approaches heavily depend on the radiometric properties of the features and only incorporate geometry information to improve performance and stability. The extracted radiometric descriptors of the features often assume a local planar or smooth object, which is by definition neither present at object corners nor edges. Therefore it would be desirable to use only descriptors that are rigorously founded for the given object model. Unfortunately the task of feature matching based on radiometric properties becomes extremely difficult for this much weaker descriptors. Hence a key feature of the presented framework is the consistent and rigorous use of statistical properties of the extracted geometric entities in the matching process, allowing a unified algorithm for matching points and lines in multiple views using solely the geometric properties of the extracted features. The results are stabilized by the use of many images to compensate for the lack of radiometric information. Radiometric descriptors may be consistently included into the framework for stabilization as well. Results from the application of the presented framework to the task of fully automatic reconstruction of points and lines from multiple images are shown.

@INPROCEEDINGS{beder04:unified, author = {Christian Beder}, title = {A Unified Framework for the Automatic Matching of Points and Lines in Multiple Oriented Images}, booktitle = {Proc. 20th ISPRS Congress, Istanbul, Turkey}, year = {2004}, pages = {1109-1113}, organization = {ISPRS}, pdf = {http://www.ipb.uni-bonn.de/papers/2004/beder04.unified.pdf}, }

@inproceedings{förstner04:projective, author = "Förstner, Wolfgang", title = "Projective Geometry for Photogrammetric Orientation Procedures I", year = "04", city = "Bonn", proceeding = "Tutorial notes from the tutorial held at the ISPRS Congress Istanbul",}

@inproceedings{förstner04:projective, author = "Förstner, Wolfgang", title = "Projective Geometry for Photogrammetric Orientation Procedures II", year = "04", city = "Bonn", proceeding = "Tutorial notes from the tutorial held at the ISPRS Congress Istanbul",}

During the last years the number of available low-cost digital consumer cameras has significantly increased while their prices decrease. Therefore for many applications with no high-end accuracy requirements it is an important consideration whether to use low-cost cameras. This paper investigates in the use of consumer cameras for photogrammetric measurements and vision systems. An important aspect of the suitability of these cameras is their geometric stability. Two aspects should be considered: The change of calibration parameters when using the camera's features such as zoom or auto focus and the time invariance of the calibration parameters. Therefore laboratory calibrations of different cameras have been carried out at different times. The resulting calibration parameters, especially the principal distance and the principal point, and their accuracies are given. The usefulness of the information given in the image header, especially the focal length, is compared to the results of the calibration.

@inproceedings{läbe*04:geometric, author = "Läbe, Thomas and Förstner, Wolfgang", title = "Geometric Stability of Low-Cost Digital Consumer Cameras", year = "04", pages = "528-535", city = "Bonn", proceeding = "Proc. of XXth ISPRS Congress 2004",}

The paper highlights approaches to reference data acquisition in real environments for the purpose of performance evaluation of image analysis procedures. Reference data for the input and for the output of an algorithm is obtained by a) exploiting the noise characteristics of Gaussian image pyramids and b) exploiting multiple views. The approaches are employed exemplarily in the context of evaluating low level feature extraction algorithms.

@INPROCEEDINGS{luxen04:performance, author = {Luxen, M.}, editor = {M. Orhan ALTAN}, title = {Performance Evaluation in Natural and Controlled Environments applied to Feature Extraction Procedures}, booktitle = {Proceedings of the 20$^{th}$ \mbox{ISPRS} Congress}, year = {2004}, volume = {XXXV, Part B3}, number = {B3}, pages = {1061-1067}, organization = {ISPRS}, series = {{The International Archives of The Photogrammetry, Remote Sensing and Spatial Information Sciences}}, pdf = {http://www.ipb.uni-bonn.de/papers/2004/luxen04.performance.pdf}, address = {Istanbul, Turkey}, url = {http://www.ipb.uni-bonn.de/papers/2004/luxen04.performance.html}, postscript = {http://www.ipb.uni-bonn.de/papers/2004/luxen04.performance.ps.gz} }

With the increasing number of cameras the need for plug-and-play calibration procedures arises to realize a subsequent automatic geometric evaluation of observed scenes. An easy calibration procedure is proposed for a non-zooming stationary camera observing objects of initially equal and known heights above a ground plane. The image coordinates of the corresponding foot and head points of these objects serve as observations. For the interior and exterior orientation of the camera a minimal parametrization is introduced with the height of the camera above the ground plane, its pitch and roll angle and the principal distance. With the idea of corresponding foot and head trajectories being homologue, the situation can be reformulated with a virtual second camera observing the scene. Therefore a plane induced homography can be established for the observation model. This special planar homology can be parametrisied with the unknown calibration quantities. Initially the calibration is estimated by observing foot and head points of objects with known heights. In the subsequent evaluation phase the height and positions of unknown objects can be determined. With the same procedure the calibration can be checked and updated if needed. The approach is evaluated with a real scene.

@INPROCEEDINGS{meidow04:calibration, author = {Jochen Meidow}, title = {Calibration of Stationary Cameras by Observing Objects of Equal Heights on a Ground Plane}, booktitle = {Proc. 20th ISPRS Congress, Istanbul, Turkey}, year = {2004}, pages = {1067-1072}, organization = {ISPRS}, pdf = {http://www.ipb.uni-bonn.de/papers/2004/meidow04.calibration.pdf}, }

We present an investigation on the use of Tensor Voting for categorizing LIDAR data into outliers, line elements (e.g. high-voltage power lines), surface patches (e.g. roofs) and volumetric elements (e.g. vegetation). The Reconstruction of man-made objects is a main task of photogrammetry. With the increasing quality and availability of LIDAR sensors, range data is becoming more and more important. With LIDAR sensors it is possible to quickly aquire huge amounts of data. But in contrast to classical systems, where the measurement points are chosen by an operator, the data points do not explicitly correspond to meaningful points of the object, i.e. edges, corners, junctions. To extract these features it is necessary to segment the data into homogeneous regions wich can be processed afterwards. Our approach consists of a two step segmentation. The first one uses the Tensor Voting algorithm. It encodes every data point as a particle which sends out a vector field. This can be used to categorize the pointness, edgeness and surfaceness of the data points. After the categorization of the given LIDAR data points also the regions between the data points are rated. Meaningful regions like edges and junctions, given by the inherent structure of the data, are extracted. In a second step the so labeled points are merged due to a similarity constraint. This similarity constraint is based on a minimum description length principle, encoding and comparing different geometrical models. The output of this segmentation consists of non overlapping geometric objects in three dimensional space. The aproach is evaluated with some examples of Lidar data.

@INPROCEEDINGS{beder04:unified, author = {Schuster, H.-F.}, title = {Segmentation Of LIDAR Data Using The Tensor Voting Framework}, booktitle = {Proc. 20th ISPRS Congress, Istanbul, Turkey}, year = {2004}, pages = {1073-1078}, organization = {ISPRS}, }

The need of describing the quality of data ranges from data acquisition to the use of the data in geoinformation systems. The contractor should verify that the data he captured suffices the specifications and the end user wants to know, if the data is suited for a special task at hand. Both are interested in quantifying the quality, possibly by simple and meaningful measures, which can be easily computed without much further efforts ­ prohibitive with respect to involved labour and related costs. Much work has been already done on the standardization of principles of quality evaluation, reports and metadata (c.f. ISO standards 19113, 19114 and 19115), but only few contributions deal with the question of defining quality measures for a specific application, which possibly may be generalized for others as well. A recent project in cooperation with the Surveying Office of North Rhine-Westphalia investigates the topic of quality evaluation of photogrammetrically captured building models with the aim to identify useful quality measures which can be used for contract specificatios and to implement an approach for automated quality control based on a comparision of measurement and reference data. This paper presents the concept of the approach and first results.

@inproceedings{appel*03:new, author = "Appel, Mirko and Weidner, Uwe", title = "A New Approach Towards Quantative Quality Evaluation of 3D Building Models", year = "03", city = "Bonn", proceeding = "ISPRS",}

The paper discusses the notion scale within geosciences. The high complexity of the developed models and the wide range of participating disciplines goes along with different notions of scale used during data acquisition and model building. The paper collects the different notions of scale shows the close relations between the different notions: map scale, resolution, window size, averqage wavelength, level of aggregation, level of abstraction. Finally the problem of identifying scale in models is discussed. A synopsis of the continuous measures for scale links the different notions.

@inproceedings{förstner03:notions, author = "Förstner, Wolfgang", title = "Notions of Scale in Geosciences", year = "03", pages = "17-39", city = "Bonn", proceeding = "Dynamics of Multi-Scale Earth Systems",}

The paper describes the automatic learning of parameters for self-diagnosis of a system for automatic orientation of single aerial images used by the State Survey Department of Northrhine--Westfalia. The orientation is based on 3D lines as ground control features, and uses a sequence of probabilistic clustering, search and ML-estimation for robustly estimating the 6 parameters of the exterior orientation of an aerial image. The system is interpreted as a classifier, making an internal evaluation of its success. The classification is based on a number of parameters possibly relevant for self-diagnosis. A hand designed classifier reached 11% false negatives and 2% false positives on appr. 17000 images. A first version of a new classifier using support vector machines is evaluated. Based on appr. 650 images the classifier reaches 2 % false negatives and 4% false positives, indicating an increase in performance.

@inproceedings{förstner*03:learning, author = "Förstner, Wolfgang and Läbe, Thomas", title = "Learning Optimal Parameters for Self-diagnosis in a System for Automatic Exterior Orientation", year = "03", pages = "236-246", city = "Bonn", proceeding = "Computer Vision Systems (ICVS) 2003",}

The paper proposes variance component estimation (VCE) for empirical quality evaluation in computer vision. An outline is given for the scope of variance component estimation in the context of quality evaluation. The principle of variance component estimation is explained and the approach is applied to results of low level feature extraction. Ground truth is only partly needed for estimating the precision, accuracy and bias of extracted points and straight line segments. The results of diverse feature extraction modules are compared.

@InProceedings{luxen03:variance, author = {Marc Luxen}, title = {Variance Component Estimation in Performance Characteristics Applied to Feature Extraction Procedures}, pages = {498-506}, editor = {Bernd Michaelis and Gerald Krell}, title = {Pattern Recognition, 25th DAGM Symposium, Magdeburg, Germany, September 10-12, 2003, Proceedings}, booktitle = {Pattern Recognition, 25th DAGM Symposium, Magdeburg, Germany, September 10-12, 2003, Proceedings}, publisher = {Springer}, series = {Lecture Notes in Computer Science}, volume = {2781}, year = {2003}, isbn = {3-540-40861-4}, crossref = {DBLP:conf/dagm/2003}, bibsource = {DBLP, http://dblp.uni-trier.de} }

The paper gives an introduction into the problem of parameter estimation from incomplete data and presents the Expectation Maximization Algorithm as a method for solving such problems. The algorithm is put in relation to geodetic estimation problems. Its practicability is shown by an example of line extraction from digital images. Zusammenfassung Der Artikel behandelt die Auswertung von unvollständigen Beobachtungsdaten. Zur Lösung der entsprechenden Schätzaufgabe wird der Expectation Maximization (EM) - Algorithmus vorgestellt. Das Problem der Schätzung unbekannter Parameter aufgrund unvollständiger Beobachtungsdaten wird in den geodätischen Kontext eingeordnet. Anhand der Linienextraktion aus digitalen Bildern wird der EM-Algorithmus exemplarisch dargestellt.

@Article{luxen*03:parameterschaetzung, author = {Luxen, M., Brunn, A.}, title = {Parameterschätzung aus unvollständigen Beobachtungsdaten mittels des EM-Algorithmus}, journal = {Zeitschrift für Geodäsie, Geoinformation und Landmanagement (ZfV)}, year = {2003}, number = {02}, pages = {71--79} }

Mit dem Aufkommen von flächig erfaßten Entfernungsdaten im Vermessungswesen steht ein Paradigmenwechsel in der Auswertung und Verarbeitung dieser Daten an, vergleichbar dem Übergang von der analytischen zur digitalen Photogrammetrie mit der Verfügbarkeit digitaler bzw. digitalisierter Bilder. Der vorliegende Beitrag gibt einen Überblick über Verfahren zur Fusion und Segmentierung von Entfernungsdaten und verdeutlicht Potentiale zur weiteren Automatisierung

@inproceedings{schuster*03:segmentierung,, author = "Schuster, Hanns-Florian and Förstner, Wolfgang", title = "Segmentierung, Rekonstruktion undDatenfusion bei der Objekterfassung mit Entfernungsdaten - ein Überblick", year = "03", city = "Bonn", proceeding = "Proceedings 2. Oldenburger 3D-Tage",}

@MISC{foerstner02:mapping, author = {W. F{\"o}rstner}, title = {{Mapping on Demand - A Dream}}, howpublished = {http://www.vernon.eu/ECVision/research\_planning/Research\_Dreams.htm}, year = {2002}, owner = {wf}, timestamp = {2011.05.22} }

The emerging interaction between Computer Vision and Photogrammetry certainly is well in the flavor of Kennert Torlegard's professional life: Not only his PhD thesis dealt with un-calibrated cameras, not only was one of his main interests close range photogrammetry with all its various applications, no, he also was active in bringing the researchers of both fields together. This paper on one side collects experiences of the dialog between Computer Vision and Photogrammetry. On the other side it gives an example, closely related to Kennert Torlegards PhD thesis of a type of analysis hopefully useful for researchers from both fields, illuminating the common fields geometry and statistics and the possibilities of mutual exchange, and finally reflects on the recent developments in the area of cognitive vision and their relation to aerial image interpretation.

@inproceedings{förstner02:computer, author = "Förstner, Wolfgang", title = "Computer Vision and Photogrammetry --- Mutual Questions: Geometry, Statistics and Cognition", year = "02", pages = "151-164", city = "Bonn", proceeding = "Bildteknik/Image Science, Swedish Society for Photogrammetry and Remote Sensing",}

In this paper we present a new method for single image orientation using an orthographic drawing or map of the scene. Environments which are dominated by man made objects, such as industrial facilities or urban scenes, are very rich of vertical and horizontal structures. These scene constraints reflect in symbols in an associated drawing. For example, vertical lines in the scene are usually marked as points in a drawing. The resulting orientation may be used in augmented reality systems or for initiating a subsequent bundle adjustment of all available images. In this paper we propose to use such scene constraints taken from a drawing to estimate the camera orientation. We use observed vertical lines, horizontal lines, and points to estimate the projection matrix P of the image. We describe the constraints in terms of projective geometry which makes them straightforward and very transparent. In contrast to the work of Bondyfalatetal 2001, we give a direct solution for P without using the fundamental matrix between image and map as we do not need parallelity constraints between lines in a vertical plane other than for horizontal lines, nor observed perpendicular lines. We present both a direct solution for P and a statistically optimal, iterative solution, which takes the uncertainties of the contraints and the observations in the image and the drawing into account. It is a simplifying modification of the eigenvalue method of Matei/Meer 1997. The method allows to evaluate the results statistically, namely to verify the used projection model and the assumed statistical properties of the measured image and map quantities and to validate the achieved accuracy of the estimated projection matrix P. To demonstrate the feasibility of the approach, we present results of the application of our method to both synthetic data and real scenes in industrial environment. Statistical tests show the performance and prove the rigour of the new method.

@InProceedings{appel*02:scene, author = {M. Appel and W. F{\"o}rstner}, title = {Scene Constraints for Direct Single Image Orientation with Selfdiagnosis}, booktitle = {Photogrammetric Computer Vision, Graz}, year = {2002}, pages = {42--49}, editor = {F. Leberl and R. Kalliany}, volume = {A} }

Eine der notwendigen Voraussetzungen zur Erstellung eines Orthophotos ist die Kenntnis der äußeren Orientierung des zu bearbeitenden Bildes. Hierfür wurde am Institut für Photogrammetrie der Universität Bonn innerhalb eines Kooperationsprojektes mit dem Landesvermessungsamt Nordrhein-Westfalen ein vollautomatisches Verfahren entwickelt, das auf der Suche von projizierten 3D-Kanten im Bild basiert. Das Programm trägt den Namen "AMOR" (Automatische Modellgestützte ORientierung) und beinhaltet sowohl Bildverarbeitung (Kantenextraktion) als auch robuste Schätzverfahren für die Bestimmung der Orientierungselemente. Als Datenbasis zur Bestimmung der äußeren Orientierung werden anders als beim konventionellen manuellen Vorgehen keine Passpunkte sondern sogenannte "Passpunktmodelle" verwendet. Dies sind Mengen georeferenzierter 3D-Kanten, wofür sich insbesondere Gebäudekanten eines digitalen Gebäudemodells eignen. Zur fächendeckenden Orthophotoproduktion wurden in Nordrhein-Westfalen landesweit Gebäude als Passpunktmodelle erfasst. AMOR ist in den Produktionsablauf der Orthophotoherstellung beim Landesvermessungsamt integriert worden und kann aufgrund der Passpunktmodelldatenbank auf einem Großteil der Landesfläche angewendet werden. Der Aufsatz gibt einen Überblick über das Verfahren zur automatischen Orientierungsbestimmung und dessen Integration mit besonderem Schwerpunkt auf den praktischen Einsatz beim Landesvermessungsamt NRW.

@inproceedings{läbe*02:automatische, author = "Läbe, Thomas and Henze, Manfred", title = "Automatische äussere Orientierung in der Orthophotoproduktion - ein Erfahrungsbericht", year = "02", pages = "245-252", city = "Bonn", proceeding = "Proceedings of DGPF Conference",}

The paper describes a method for blind estimation of sharpness and resolving power from a single image. These measures can be used to characterize images in the context of the performance of image analysis procedures. The method assumes the point spread function (PSF) can be approximated by an anisotropic Gaussian. The width Sigma of the PSF is determined by the ratio Sigma_g/Sigma_g' of the standard deviations of the intensity and of its derivative at edges. The contrast sensitivity function (CSF) is based on an optimal model for detecting straight edges between homogeneous regions in noisy images. It depends on the signal to noise ratio and is linear in the frequency. The method is applied to artificial and real images proving that it gives valuable results.

@inproceedings{luxen*02:blind, AUTHOR = "Luxen, M. and F{\"o}rstner, W.", TITLE = "Characterizing Image Quality: Blind Estimation of the Point Spread Function from a Single Image", BOOKTITLE = "PCV02", YEAR = "2002", PAGES = "A: 205"}

The paper presents a new technique for optimal estimation for statistically uncertain geometric entites. It is an extension of the classical eigenvector solution technique but takes the full covariance information into account to arrive at a ML-estimate. The proposed solution is significantly more transparent than the solution for estimation under heteroscedasticity proposed by Leedan, Matei and Meer. We give a new representation of algebraic projective geometry easing statistical reasoning. We show how the setup can be used in object reconstruction, especially when estimating points and edges of polyhedra. We explicitely give an example for estimating 3D-points and 3D-lines from image points and image lines. The direct solutions do practically require no approximate values.

@inproceedings{förstner01:algebraic, author = "Förstner, Wolfgang", title = "Algebraic Projective Geometry and Direct Optimal Estimation of Geometric Entities", year = "01", city = "Bonn", proceeding = "appeared at the OeAGM 2001",}

Orientation of cameras with minimum or redundant information is the first step in 3D-scene analysis. The difficulty of this task lies in the lack of generic and robust procedures for geometric reasoning, calibration and especially orientation. The paper collects available tools from statistics, expecially for the diagnosis of data and design and for coping with outliers using robust estimation tecniques. It presents a generic strategy for data analysis on the contest of orientation procedures which may be extendet towards self-calibration.

@INCOLLECTION{foerstner01:generic, author = {F{\"o}rstner, W.}, editor = {A. Gruen and T. S. Huang}, year = 2001, booktitle = {Calibration and Orientation of Cameras in Computer Vision}, title = {Generic Estimation Procedures for Orientation with Minimum and Redundant Information}, number = {34}, series = {Series in Information Sciences}, publisher = {Springer} }

The paper presents tools for optimally estimating 3D points and lines from 2D points and lines. It uses algebraic projective geometry for representing 2D and 3D geometric entities, perspective projection and its inversion. The uncertainty of the entities can easily be integrated. The direct solutions do not require approximate values.

@InCollection{foerstner01:estimating, author = {F{\"o}rstner , W.}, title = {{On Estimating 2D Points and Lines from 2D Points and Lines}}, booktitle = {Festschrift anl{\"a}\ss lich des 60. Geburtstages von Prof. Dr.-Ing. Bernhard Wrobel}, pages = {69 -- 87}, publisher = {Technische Universit{\"a}t Darmstadt}, year = 2001 }

We present a method for estimating unknown geometric entities based on identical, incident, parallel or orthogonal observed entities. These entities can be points and lines in 2D and points, lines and planes in 3D. We don't need any approximate values for the unknowns. The entities are represented as homogeneous vectors or matrices, which leads to an easy formulation for a linear estimation model. Applications of the estimation method are manifold, ranging from 2D corner detection to 3D grouping.

@InProceedings{heuel01:points, author = {Heuel, S.}, title = {Points, Lines and Planes and their Optimal Estimation}, booktitle = {Pattern Recognition, 23rd DAGM Symposium}, year = 2001, pages = {92-99}, number = 2191, series = {LNCS}, month = {September}, publisher = {Springer}, url = {http://www.ipb.uni-bonn.de/ipb/lit/papers01/heuel01.points.html}, postscript = {http://www.ipb.uni-bonn.de/ipb/lit/papers01/heuel01.points.ps.gz}, }

Automatic building extraction from images is a particularly hard object recognition problem, because both the image data and the models to be reconstructed reveal a high complexity. Whereas models have to be generic in order to fit most of the observable different building shapes they also have to be building specific to discriminate buildings from other objects in the images. This situation describes the typical dilemma that all building recognition system have to cope with. In this article we present and discuss two approaches for automatic building reconstruction that were developed at the University of Bonn during the last eight years. It is shown how different AI methods were employed to solve the numerous problems concerning modeling, inference and uncertain reasoning, matching, and evaluation. Both approaches have been implemented and were successfully applied to real data. Due to complementary limitations, the integration of both would be desirable in order to develop a more comprehensive solution. Unfortunately, a number of open questions still have to be answered which are discussed at the end of the article.

@Article{heuel*01:building, author = {Heuel, S. and Kolbe, T.H.}, title = {{Building {R}econstruction: {T}he {D}ilemma of {}Generic {V}ersus {S}pecific {M}odels}}, journal = {KI -K{\"u}nstliche Intelligenz}, number = 3, pages = {57-62}, month = {July}, url = {http://www.ipb.uni-bonn.de/papers/2001/heuel01.building.html} year = 2001, }

We present a geometric method for (i) matching 2D line segments from multiple oriented images, (ii) optimally reconstructing 3D line segments and (iii) grouping 3D line segments to corners. The proposed algorithm uses two developments in combining projective geometry and statistics, which are described in this article: (i) the geometric entities points, lines and planes in 2D and 3D and their uncertainty are represented in homogeneous coordinates and new entities may be constructed including their propagated uncertainty. The construction can be performed directly or as an estimation. (ii) relations such as incidence, equality, parallelity and orthogonality between points, lines and planes can be tested statistically based on a given significance level. Using these tools, the resulting algorithm is straight-forward and gives reasonable results. It is only based on geometric information and does not use any image intensities, though it can be extended to use other information. The matching of 3D lines does not need any thresholds other than a significance value for the hypotheses tests.

@InProceedings{heuel*01:matching, author = {Heuel, S. and F{\"o}rstner, W.}, title = {Matching, Reconstructing and Grouping 3D Lines From Multiple Views Using Uncertain Projective Geometry}, booktitle = {CVPR '01}, year = 2001, organization = {IEEE}, url = {http://www.ipb.uni-bonn.de/ipb/lit/papers01/heuel01.matching.html}, postscript = {http://www.ipb.uni-bonn.de/ipb/lit/papers01/heuel01.matching.ps.gz}, note = {accepted for publication} }

The paper discusses models for building extraction from multiple images and shows the importance of the joint use of topological relations and uncertain geometry resulting in a platform for spatial reasoning useful for the reconstruction of manmade objects. We motivate our approach based on the experience in building reconstruction and describe tools for topological and geometric reasoning under uncertainty. We use a polyhedral patch model as intermediate layer for building interpretation.

@InProceedings{heuel*01:topological, title = {Topological and geometrical models for building extraction from multiple images}, author = {Heuel, S. and F{\"o}rstner, W.}, booktitle = {Automatic Extraction of Man-Made Objects from Aerial and Space Images (III)}, year = 2001, url = {http://www.ipb.uni-bonn.de/ipb/lit/papers01/heuel01.topological.html}, publisher = {Balkema Publishers} }

The paper presents an optimal estimate for the projection matrix for points of a camera from an arbitrary mixture of six or more observed points and straight lines in object space. It gives expressions for determining the corresponding projection matrix for straight lines together with its covariance matrix. Examples on synthetic and real images demonstrate the feasibility of the approach.

@inproceedings{luxen*01:optimal, author = "Luxen, Marc and Förstner, Wolfgang", title = "Optimal Camera Orientation from Points and Straight Lines", year = "01", pages = "84-91", city = "Bonn", proceeding = "Proceedings of the DAGM 2001",}

For optical 3D reconstruction, specially for real-time image sequence calculations, highly efficient algorithms are required. We discuss two aspects of increasing the efficiency of a matching algorithm based on feature points. The first aspect is the efficiency of checking the consistency of matching candidates using epipolar and trifocal constraints. The second aspect, namely the possibility of approximating the non projective mapping of multi-media geometry by a projective one, which leads to virtual cameras, is investigated in the main part of the paper. Exploiting the simplicity of algebraic expressions using normalized projective cameras we significantly increase the efficiency of the geometric computation during multiple image matching.

In this paper we describe a new concept for the reconstruction of buildings. In contrast to most of the published approaches, we link the reconstruction process with the building interpretation. With this linkage we want to enhance the reconstruction result and to yield semantic information about the buildings. We introduce building models based on their topology. We also may use data from different sensor types. The analysis is done locally using statistical building information for the interpretation in a Markov-Random-Field and using e. g. geometric or radiometric "appearance" models for the reconstruction. A real data example from laserscanner observations demonstrates the approach. Zusammenfassung In diesem Artikel beschreiben wir ein neues Verfahren zur Rekonstruktion von Gebauden. Im Gegensatz zu den meisten in der Literatur bereits veroffentlichen Verfahren, verbinden wir die Rekonstruktion mit der Interpretation. Dadurch ¨ verbessern wir das Ergebnis der Rekonstruktion und erhalten zusatzlich semantische Information uber das Gebaude. Wir ¨ ¨ ¨ verwenden ein Gebaudemodell, das auf der Topologie der Gebaude definiert ist. Auserdem ist die Integration von Daten ¨ ¨ ¨ unterschiedlicher Sensortypen moglich. Die Analyse der Daten erfolgt bei der Interpretation mit lokalem statistischen ¨ Gebaudewissen und bei der Rekonstruktion mit lokalen Modellen der "Erscheinungsform" der Gebaude (z. B. geometrisch ¨ ¨ oder radiometrisch). Ein Beispiel mit realen Entfernungsdaten demonstriert den Ansatz.

@InProceedings{brunn00:step, author = "A. Brunn", title = " A step towards semantic-based Building Reconstruction Using Markov-Random-Fields", booktitle = "ISPRS Congress, Amsterdam", year = 2000 }

The paper discusses models for building extraction from multiple images and shows the importance of the joint use of topological relations and uncertain geometry resulting in a platform for spatial reasoning useful for the reconstruction of manmade objects. We motivate our approach based on the experience in building reconstruction and describe tools for topological and geometric reasoning under uncertainty. We use a polyhedral patch model as intermediate layer for building interpretation.

@Article{brunn00:analyse, author = "A. Brunn", title = "Analyse von Laserscannerdaten zur Geb{\"a}uderekonstruktion", journal = "Zeitschrift f{\"u}r Photogrammetrie, Fernerkundung und Geoinformationssysteme", year = 2000, pages = "189-190", number = 3}

The paper discusses preprocessing for feature extraction in digital intensity, color and range images. Starting from a noise model, we develop estimates for a signal dependent noise variance function and a method to transform the image, to achieve an image with signal independent noise. Establishing significance tests and the fusion of different channels for extracting linear features is shown to be simplified.

@incollection{foerstner00:image, author = {W. F{\"o}rstner}, title = {{Image Preprocessing for Feature Extraction in Digital Intensity, Color and Range Images}}, booktitle = {{Geomatic Methods for the Analysis of Data in Earth Sciences}}, volume = {95/2000}, series = {{Lecture Notes in Earth Sciences}}, pages = {165--189}, publisher = {Springer}, year = {2000} }

Optimally reconstructing the geometry of image triplets from point correspondences requires a proper weighting or selection of the used constraints between observedcoordinates andunknown parameters. By analysing the ML-estimation process the paper solves a set of yet unsolved problems: (1) The minimal set of four linearily independent trilinearities (Shashua 1995, Hartley 1995) actually imposes only three constraints onto the geometry of the image triplet. The seeming contradiction between the number of used constraints, three vs. four, can be explained naturally using the normal equations. (2) Direct application such an estimation suggests a pseudoinverse of a 4 × 4-matix having rank 3 which contains the covariance matrix of the homologeous image points to be the optimal weight matrix. (3) Insteadof using this singluar weight matrix one could select three linearily dependent constraints. This is discussed for the two classical cases of forward and lateral motion, and clarifies the algebraic analyis of dependencies between trilinear constraints by Faugeras 1995. Results of an image sequence with 800 images and an Euclidean parametrization of the trifocal tensor demonstrate the feasibility of the approach.

@inproceedings{förstner00:optimally, author = "Förstner, Wolfgang", title = "Optimally Reconstructing the Geometry of Image Triplets", year = "00", pages = "669-684", city = "Bonn", proceeding = "Appeared in: Computer Vision - ECCV 2000",}

@inproceedings{förstner00:new, author = "Förstner, Wolfgang", title = "New Orientation Procedures", year = "00", pages = "3A 297-304", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

This paper integrates statistical reasoning and Grassmann-Cayley algebra for making 2D and 3D geometric reasoning practical. The multi-linearity of the forms allows rigorous error propagation and statistical testing of geometric relations. This is achieved by representing all objects in homogeneous coordinates and expressing all relations using standard matrix calculus.

@INPROCEEDINGS{foerstner*00:statistically, author = {F{\"o}rstner, W. and Brunn, A. and Heuel, S.}, title = {Statistically Testing Uncertain Geometric Relations}, booktitle = {Mustererkennung 2000}, pages = {17-26}, year = 2000, editor = {Sommer,G. and Kr{\"u}ger, N. and Perwass, Ch.}, month = {September}, organization = {DAGM}, publisher = {Springer}, url = {http://www.ipb.uni-bonn.de/papers/2000/foerstner00.testing.html}, postscript = {http://www.ipb.uni-bonn.de/papers/2000/foerstner00.testing.ps.gz} }

@inproceedings{förstner*00:exploiting, author = "Förstner, Wolfgang and Wolff, Kirsten", title = "Exploiting the Multi View Geometry for Automatic Surfaces Reconstruction Using Feature Based Matching in Multi Media Photogrammetry", year = "00", pages = "5B 900-907", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

The objective of the presented work is the automatic segmentation of urban areas from high resolution satellite images, such as MOMS-02 images or from aerial images taken from high altitude flights. The structure of urban areas, as seen from satellites or aeroplanes, is mainly characterized by three elements: the road network, the morphology of the built up areas and the distribution of the vegetation. There exist many types of road structures in large cities, which govern the local topology and geometry of the individual roads. Typical examples are orthogonal networks, star type networks or irregular networks. Seen world wide, orthogonal networks appear to be the most common ones, as e. g. to be found in Mannheim, Barcelona, New York or Canberra. The paper presents an approach for segmentation of dominant orthogonal road structures from high resolution satellite images, like MOMS-02, or aerial images.

@inproceedings{faber*00:detection, author = "Faber, Anette and Förstner, Wolfgang", title = "Detection of dominant orthogonal road structures in small scale", year = "00", pages = "3A 274-281", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

In our paper we describe the ongoing research to develop an intelligent airbag system. Using image sequences acquired from a stereo camera, we detect the form and the position of the driver and passenger seat. And, if a seat is classified as occupied, we try to estimate the geometry and position of the human's head as the most distinguishing feature of the body, if it is possible. The developed software system consists of five steps: the correction of distortions followed by an epipolar rectification of the stereo images, the feature extraction, the feature-based matching, the seat occupation detection and verification, and the approximation of the human's head. The emphasis in this paper is on the used model to approximate the human's head by an ellipsoid. The base of a subset of the estimated features as well as a certain assignment of the features to a human's head. The used model bases on the least square method with a condition, which supports the approximation of an ellipsoid. To determine, if an obtained approximation is valid the result is compared with the standard dimensions of a human's head. On tests on numerous image sequences recorded inside different vehicles the feasibility of the system is shown. The information about the seat occupation and the location of the detected passengers inside the vehicle can be used to control an intelligent airbag deployment. 

@inproceedings{faber00:image-based, author = "Faber, Petko", title = "Image-based passenger detection and localization inside vehicles", year = "00", pages = "5A 230-238", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

In our paper we address to the problem of robust seat occupation detection inside vehicles. The used approach consists of four steps: correction of distortions followed by an epipolar rectification of the stereo images, feature extraction, feature-based matching, and the seat occupation detection and verification. The focus in this paper is on the verification of the seat occupation. The step of verification corresponds to a classification of the driver and the passenger seat as occupied or empty. First, we try to estimate the seat geometry and localization. Implicitly it can be deduced from the results, that if a seat can be modeled adapted to the data, the seat is empty. Otherwise we can assume that the seat is occupied by an object. Then, we try to differ between an occupation by a human, or any other object. On tests on numerous image sequences recorded inside different vehicles the feasibility of the approach is shown.

@inproceedings{heuel*00:topological, author = "Heuel, Stephan and Förstner, Wolfgang and Lang, Felicitas", title = "Topological and Geometrical Reasoning in 3D Grouping for Reconstructing Polyhedral Surfaces", year = "00", pages = "3A 397-404", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

@inproceedings{ragia00:quality, author = "Ragia, Lemonia", title = "A Quality Model for Spatial Objects", year = "00", pages = "4C 855-862", city = "Bonn", proceeding = "Appeared at the Proceedings of the 19th ISPRS Congress",}

3D-city models are becoming an important tool for town planning. Photogrammetry appears to provide the only economic means to acquire truly 3D city data. The paper discusses the current status of automatic and semiautomatic acquisition method. Research in automatic methods for building extraction being increasingly intensive in the last few years has lead to promising results, however, still lacks the performance needed for practical applications. Semiautomatic acquisition methods rely on the ability of the operator to intelligently interpret and select the required information and appear to be ripe for practical implementation.

@inproceedings{förstner99:3d-city, author = "Förstner, Wolfgang", title = "3D-City Models: Automatic and Semiautomatic Acquisition Methods", year = "99", city = "Bonn", proceeding = "Photogrammetrische Woche",}

The paper collects tools for estimating rotations. Starting from the classical representations with quaternions concatenation rules for the Rodriguez parameters are given. Direct estimates for mean rotations and for rotations from homologous spatial directions are given. Two robust estimation procedures are given for estimating the rotation matrix of a single camera from observed straight line segments in a legoland scene based on a grouping procedure for line segment and and a clustering procedure on the 3-sphere.

@inproceedings{förstner99:estimating, author = "Förstner, Wolfgang", title = "On Estimating Rotations", year = "99", city = "Bonn", proceeding = "Festschrift für Prof. Dr.-Ing. Heinrich Ebner zum 60. Geburtstag",}

Voronoi diagrams are a classical tool for analyzing spatial neighborhood relations. For point fields the spatial proximity can be easily visualized by the dual graph, the Delaunay triangulation. In image analysis VDs and DTs are commonly used to derive neighborhoods for grouping or for relational matching. Neighborhood relations derived from the VD, however, are uncertain in case the common side of two Voronoi cells is comparably short or, equivalently, in case four points of two neighboring triangles in a DT are close to a circle. We propose a measure for characterizing the uncertainty of neighborhoods in a plane point field. As a side result we show the measure to be invariant to the numbering of the four points, though being dependent on the cross ratio of four points. Defining a fuzzy Delaunay triangulation is taken as an example.

@inproceedings{förstner99:uncertain, author = "Förstner, Wolfgang", title = "Uncertain Neighborhood Relations of Point Sets and Fuzzy Delaunay Triangulation", year = "99", city = "Bonn", proceeding = "Proceedings of DAGM Symposium Mustererkennung",}

The paper presents a metric for positive definite covariance matrices. It is a natural expression involving traces and joint eigenvalues of the matrices. It is shown to be the distance coming from a canonical invariant Riemannian metric on the space $Sym^+(n,\mR)$ of real symmetric positive definite matrices In contrast to known measures, collected e.~g. in Grafarend 1972, the metric is invariant under affine transformations and inversion. It can be used for evaluating covariance matrices or for optimization of measurement designs. Keywords: Covariance matrices, metric, Lie groups, Riemannian manifolds, exponential mapping, symmetric spaces

@inproceedings{förstner*99:metric, author = "Förstner, Wolfgang and Moonen, Boudewijn", title = "A Metric for Covariance Matrices", year = "99", city = "Bonn", proceeding = "Quo vadis geodesia ...?, Festschrift for Erik W. Grafarend on the occasion of his 60th birthday",}

We want to extract urban structures from aerial images or high resolution satellite scenes like German MOMS02 sensor. We aim at a separation of neighbored textured regions importend for describing different urban structures, however, do not intend to reach a full partitioning of the image. It has been shown, that grey level segmentation alone is not sufficient to approach this problem. Because some kinds of segmentation errors, like over- or undersegmentation, are possible. If we want to use texture additionally, the development of a suitable representation of texture images is required. We use the scale characteristics of the local autocovariance function, called SCAF, of the possibly multiband image function. This characteristics is derived in two steps: * The parameters strength, direction and anisotropy of the texture are derived from the squared gradient of the image function. * The spatial frequency content of these features is then determined using a Laplacian pyramid, which separates the different spectral bands of the texture. The process goes on with analyzing eigenvalues of the negative Hessian as an approximation of autocovariance function. So we obtain the three described texture parameters strength, direction and anisotropy. For these parameters we then compute the Laplace pyramid representing the spectral decomposition of the signal. For processing textured images in the described way, we fuse all image channels by integrating the squared gradient of the texture features in all channels of the image pyramid again. The normalization uses the theoretical noise behavior of the filter kernels of the Laplace pyramid, which we determined by analyzing the impulse response. The final result of the process then are the texture edges. The paper contains a presentation of our method and results for artificial and natural scenes.

In diesem Beitrag wird ein Ansatz zur Segmentierung einer aus einer ikonischen Beschreibung eines Bildes extrahierten Kontur in eine Menge von Kurven verschiedener Ordnungen vorgestellt. Die Funktion des allgemeinen Ansatzes wird anhand der Zerlegung einer Kontur in eine Menge von Geraden- und Ellipsensegmente gezeigt. Charakteristische Eigenschaften dieses Ansatz sind: (1) die parameterlose Segmentierung, (2) die Invarianz gegenüber geometrischen Transformationen, (3) die Simplizität des gewühlten Ansatzes und (4) die verhältnismäßig effiziente Segmentierung.

@inproceedings{faber99:parameterlose, author = "Faber, Petko", title = "Parameterlose Kontursegmentierung", year = "99", pages = "172-180", city = "Bonn", proceeding = "Proceedings of DAGM Symposium Mustererkennung",}

There is a considerable interest in the computer vision and computer graphics community to generate a suitable symbolic description of 2D curves. This paper presents a new approach to generate a compact description by a nonparametric segmentation algorithm. Important is that no thresholds are required to determine the segmentation, which best describes a 2D curve. The result is a symbolic description by a set of features of different order. The emphasis is on the used significance measure referring to the limits of an acceptable interpretableness of the reachable sementation results. the proposed algorithm has a number of interesting properties: (1) independence of the segmentation from any parameters, (2) invariance to geometric transformations, (3) simplicity, and (4) efficiency of the segmentation algorithm.

@inproceedings{faber99:nonparametric, author = "Faber, Petko", title = "Nonparametric segmentation of 2D curves", year = "99", city = "Bonn", proceeding = "Institut für Photogrammetrie",}

@inproceedings{fischer*99:use, author = "Fischer, André and Kolbe, Thomas H. and Lang, Felicitas", title = "On the Use of Geometric and Semantic Models for Component-Based Building Reconstruction", year = "99", pages = "101-119", city = "Bonn", proceeding = "Proceedings: Semantic Modeling for the Acquisition of Topographic Information from Images and Maps, Smati 99 Workshop",}

The modeling of three-dimensional objects is a current topic in digital photogrammetric research. The modeling of buildings in digital imagery or digital surface models involving automation processes has reached a level where it can compete with classical photogrammetric stereo measurements. There are many different ways on how to integrate automation. We describe our system and its automated features that support the operator in the adaption of parametric models to multiple overlapping images. There do exist tools to automate the measurement of heights, to automate the estimation of the form parameters or for the handling of building aggregates. With such tools we can reach about 20 seconds for the modeling of a volumetric primitive which is fully comparable to the currently used photogrammetric methods.

@inproceedings{gülch*99:integration, author = "Gülch, Eberhard and Müller, Hardo and Läbe, Thomas", title = "Integration of Automatic Processes Into Semi-Automatic Building Extraction", year = "99", city = "Bonn", proceeding = "Proceedings of ISPRS Conference "Automatic Extraction Of GIS Objects From Digital Imagery"",}

We are developing a system for reconstructing polyhedral surfaces from multiple images. This process can take advantage of the topological relations of the observed image features triggering and therefore speeding up the grouping processes. Exploiting the statistical properties of features when grouping them leads to consistent decisions being invariant to numbering and choice of coordinate system and allows statistical testing. This simplifies the choice of thresholds to the definition of a scene independent significance level. We decribe the topological and statistical models used within our system. Experiments with synthetic and real data prove the feasibility of the approach.

This paper describes the contribution of the University of Bonn to the OEEPE test on Automatic Orientation of Aerial Images (Task A). A program for the automatic exterior orientation called AMOR was developed by Wolfgang Schickler at the Institute of Photogrammetry, Bonn. The methods and ideas of this approach are summarized. This program was used to compute the exterior orientation parameters of the given two test images successfully. Results and new solved problems are reported.

@inproceedings{läbe99:contribution, author = "Läbe, Thomas", title = "Contribution to the OEEPE-Test on Automatic Orientation of Aerial Images, Task A - Experiences with AMOR", year = "99", city = "Bonn", proceeding = "OEEPE Seminar on Automatic Orientation of Aerial Images on Database Information",}

The paper develops an approach for assessing the quality of ground plans of buildings. Quality is measured not only by geometrical but also by structural differences between an acquired data set and a reference data set. New hybrid techniques for automatically determining quality measures are developed, and shown to be applicable to real data. The uncertainty of the given data is taken into account. Automating quality assessment increases efficiency in checking data, allowing complete checks instead of sampling, moreover it makes quality checks objective. The developped techniques are applicable to sets of 2D regions of any type and internal structure. We also demonstrate the necessity to use the quality of the quality parameters when checking the fullfillment of quality specifications.

@inproceedings{ragia*99:automatically, author = "Ragia, Lemonia and Förstner, Wolfgang", title = "Automatically Assessing the Geometric and Structural Quality of Building Ground Plans", year = "99", city = "Bonn", proceeding = "Proceedings of ISPRS Working Group II/6, International Workshop on 3D Geospatial Data Production: Meeting Application requirements",}

In this paper we investigate the relation between camera calibration and structure from motion. A method is presented to analyze the effect of systematic errors and uncertainty in camera calibration on 3D-reconstruction and motion parameters. In two simple examples from stereo with lateral and forward motion the approach is demonstrated. The approach can easily be extended to more complex situations and used for planning and online diagnostics in calibration and structure from motion.

@inproceedings{abraham*98:calibration, author = "Abraham, Steffen and Förstner, Wolfgang", title = "Calibration Errors in Structure from Motion", year = "98", pages = "117-124", city = "Bonn", proceeding = "Mustererkennung 1998, 20. DAGM-Symposium",}

The paper presents methods for sensitivity analysis applied to the relation between camera calibration and structure from motion. The uncertainty of the calibration is represented by the bias and the covariance matrix of the calibration parameters, describing the effect of incomplete modeling and random errors during calibration. The effect of calibration errors onto the 3D structure is analyzed for stereo in lateral and forward motion. The results reveal interesting relations between stability and sensitivity and demonstrate the need for a rigorous statistical analysis which takes into account all mutual stochastical dependencies. As a side result, the comparison of two different calibration models, namely Tsai's model and a new polynomial model, demonstrates the limitations of Tsai's model.

This paper presents a hybrid concept of interaction between scene and sensors for image interpretation. We present a strategy for 3D building acquisition which combines different approaches based on different levels of description and different sensors: the detection of regions of interest, and the automatic and semiautomatic reconstruction of object parts and complete buildings.

@inproceedings{brunn*98:hybrid, author = "Brunn, Ansgar and Lang, Felicitas and Gülch, Eberhard and Förstner, Wolfgang", title = "A Hybrid concept for 3D Building Acquisition", year = "98", pages = "119-129", city = "Bonn", proceeding = "Journal for Photogrammetry & Remote Sensing Vol. 53",}

@inproceedings{förstner98:working, author = "Förstner, Wolfgang", title = "Working Group Report "Specification and propagation of uncertainty"", year = "98", city = "Bonn", proceeding = "(interne Veröffentlichung 1998)",}

@inproceedings{förstner98:theoretical, author = "Förstner, Wolfgang", title = "On the Theoretical Accuracy of Multi Image Matching, Restoration and Triangulation", year = "98", city = "Bonn", proceeding = "Festschrift zum 65. Geburtstag von Prof. Dr.-Ing. mult. G. Konecny",}

The paper analyses the theoretical precision of integrated multiple image matching and image reconstruction, and the theoretical accuracy of the triangulation from a sequence of images specializing to tri- and binocular stereo. The estimated geometric parameters from multi image matching, used in aerial triangulation for point transfer, turns out to be statistically uncorrelated from the restored image, and the precision of the shift between two images does not depend on the number of images taking part in the multi image matching. Triangulating from an image sequence reveals the variance of the position of points perpendicular to the trajectory to decrease with the number of images whereas the variance of the distance of the 3D-point to the trajectory decreases with the cube of the number of images, taking the distance between the images as given. The case of three images, representative for three line cameras shows the distance to be independent of the central ray.

@inproceedings{förstner98:theoretical, author = "Förstner, Wolfgang", title = "On the Theoretical Accuracy of Multi Image Matching, Restoration and Triangulation", year = "98", city = "Bonn", proceeding = "1998",}

@inproceedings{fischer*98:extracting, author = "Fischer, André and Kolbe, Thomas H. and Lang, Felicitas and Cremers, Armin B. and Förstner, Wolfgang and Plümer, Lutz and Steinhage, Volker", title = "Extracting Buildings from Aerial Images Using Hierarchical Aggregation in 2D and 3D", year = "98", city = "Bonn", proceeding = "Computer Vision and Image Understanding",}

Der Übergang von der klassischen Photogrammetrie mit analogen Bildern zur Digitalen Photogrammetrie mit digitalen Bildern und der Automation von Messprozessen eröffnet vielfältige Möglichkeiten zur effizienten Erfassung von 3D-Information in urbanen Gebieten. Es ist ein zunehmender Bedarf an realen 3D-Daten für eine Vielfalt von Anwendungen zu beobachten. Diese reichen von Managementaufgaben in Stadtplanung und Städtebau bis hin zu Funknetzsimulationen, Werbung und Spieleindustrie. 3D-Stadtmodelle eignen sich nicht nur zur Analyse und Simulation, sondern auch zur computeranimierten Darstellung bis hin zur Virtuellen Realität. Das hier vorgestellte System wird zur geometrischen Modellierung aus Mono-, Stereo- oder mehrfachüberdeckenden Luftbildern eingesetzt. Verschiedene automatisierte Module unterstützen den Operateur. Neben einer detaillierten geometrischen Erfassung spielt die Visualisierung von Texturen eine immer größere Rolle. Mit dem hier vorgestellten System können orientierte Luft- und Nahbereichsaufnahmen gleichermaßen zur automatischen Texturierung von erfaßten 3D-Objektmodellen eingesetzt werden. Integraler Bestandteil ist eine Entzerrung und die Wahl der Texturelementgröße im Objektraum, womit verschiedene Detaillierungsgrade (Level-of-Detail) der virtuellen Darstellung erzeugt werden können.

@inproceedings{gülch*98:virtuelle, author = "Gülch, Eberhard and Müller, Hardo", title = "Virtuelle Staedte aus digitalen Bildern", year = "98", city = "Bonn", proceeding = "Proceedings of Virtual GIS",}

A Semi-Automatic Building Extraction system using two or more digitized overlapping aerial images has been enhanced by increased automation for the measurement of saddle-back-roof buildings, hip-roof buildings and boxes. All newly developed modules have been incorporated in the object oriented design of the system. The new methods consist of a ground-point and roof-top matching tool and a robust determination of shape parameters, like e.g. gutter length and width. The current performance of building extraction is quantitatively and qualitatively evaluated. We examine the increased efficiency using the automated tools, the success rate of individual modules and the overall success rate using a combination of methods. A methodology for quantitative comparison is tested on footprints of buildings from classical stereo measurements and from semi-automatic measurements. A qualitative comparison in 3D of multiple measurements of complete buildings is performed on three different datasets.

@inproceedings{gülch*98:performance, author = "Gülch, Eberhard and Müller, Hardo and Läbe, Thomas and Ragia, Lemonia", title = "On the performance of semi-automatic building extraction", year = "98", city = "Bonn", proceeding = "Proceedings of ISPRS Commission III Symposium",}

Abstract In the field of reverse engineering the methods of stripe projection are predestined to record smooth homogeneous object surfaces. A disadvantage of these methods is the low level of accuracy. Photogrammetric techniques, on the other hand, offer highly accurate methods for orientation, calibration and object reconstruction. For a more efficient and accurate surface recording we propose a fusion of both techniques: photogrammetry and methods of stripe projection. This combination includes an automated orientation and calibration procedure for a stripe sensor which works according to the principle of coded light. The paper focusses on the automatic determination of approximate values and the structure of the calibration process.

@inproceedings{hau98:orientierungund, author = "Hau, Thomas", title = "Orientierungund Kalibrierung eines Rangesensors unter Nutzung geometrischer Objektinformation", year = "98", city = "Bonn", proceeding = "18. Wissenschaftlich- Technische Jahrestagung der Deutschen Gesellschaft für Photogrammetrie und Fernerkundung",}

We propose a new representation for segmented images useful for Perceptual Organization. The representation has four distinct properties: (1) It is topologically consistent,~i.e. the image plane is completely described; (2) the representation treats fore- and background symmetrically, a change of fore- and background has a well-defined and transparent impact on the representation; (3) the hierarchical structure of the representation explicitly reflects the aggregation of parts and objects; (4) finally the representation has an associated scale, which refers to the significance of image parts and of their relationships. We present an example for such a representation, where the images consist of area type features and the significance of the relationships of the blobs are based on their proximity.

@inproceedings{heuel*98:dual,, author = "Heuel, Stephan and Förstner, Wolfgang", title = "A Dual, Scalable and Hierarchical Representation for Perceptual Organization of Binary Images", year = "98", city = "Bonn", proceeding = "IEEE Workshop on Perceptual Organization in Computer Vision 1998",}

A semi-automatic building extraction system using two or more digitized overlapping aerial images has been enhanced by increased automation for the measurement of saddleback-roof (lopsided and symmetric) buildings, hip-roof buildings and flat-roof building (boxes). The goal is to minimize the interaction an operator has to do for measuring the form and pose parameters of 3D building models of the above mentioned types. The automated tasks are computed on-line and fully integrated in the work flow. Thus accepting or correcting the results or adapting the automated calculation is possible. The used methods are grey value correlation for absolute heights and the robust estimation techniques RANSAC and Clustering for the determination of heights and the other form parameters of the building primitives. These methods work on automatically extracted line segments. The automated modules have been empirically evaluated on more than 250 buildings in two datasets with different image quality and different densities of built-up areas. The results of these tests show a success rate of up to 88% for a form parameter estimation module and the height measurement.

@inproceedings{läbe*98:robust, author = "Läbe, Thomas and Gülch, Eberhard", title = "Robust Techniques for Estimating Parameters of 3D Building Primitives", year = "98", city = "Bonn", proceeding = "Proceedings of ISPRS Commission II Symposium",}

A semi-automatic system for extracting topographic features is being migrated to an object-oriented design for better maintainability. For that purpose an object-oreinted model of the extracted objects and the extraction methods is required. Moreover an appropriate software component model for the exchange with other systems is needed. We have modeled a class hierarchy for objects, that can be semi-automatically extracted from digital images. These objects are characterized by geometric, textural and reflectance properties. We have classified the extraction methods and modeled the message transfer of an interactive extraction mehtod. The component technologies CORBA and JavaBeans (TM) were used to make the extracted objects and system components available for other systems. We found out that an access to the objcets of the Semi-Automatic System by operation calls allows a more flexible data transfer and control of the system than standard file transfer. Therefore Geographic Information Systems (GIS) should support appropriate software component models to co-operate optimally with interactive or semi-automatic feature extraction systems.

@inproceedings{müller98:object-oriented, author = "Müller, Hardo", title = "Object-oriented modeling for the extraction of geometry, texture and reflectance from digital images", year = "98", city = "Bonn", proceeding = "Proceedings of ISPRS Commission IV Symposium",}

@inproceedings{moonen98:plückerkoordinaten, author = "Moonen, Boudewijn", title = "Plückerkoordinaten I: Die klassische Beschreibung", year = "98", city = "Bonn", proceeding = "(interne Veröffentlichung 1998)",}

For automated image evaluation a precise characterization of the noise behaviour of the image data is needed. Digital image acquisition in aerial photogrammetry usually is performed by digitizing analogue film material. The noise of the digital image is composed of the noise due to the digitization process and of the noise of the analogue image. From a statistical point of view, noise can be described by the standard deviation and the correlation of the fluctuation of the intensity values. Our aim is to determine a noise model out of the images itself, without the need of special test patterns. We do this by following two approaches: pixelwise evaluation of multiple scans of the film material and regionwise evaluation from a single scan. The first approach analyses the noise coming from the digitization process, while in the second case we obtain the complete noise budget. We analyze the correlation between successive scans. First empirical investigations with different film material were performed and the results are presented.

@inproceedings{waegli98:investigations, author = "Waegli, Barbara", title = "Investigations into the Noise Characteristics of Digitized Aerial Images", year = "98", city = "Bonn", proceeding = "Proceedings of ISPRS Commission II Symposium",}

In this paper we present a quality evaluation of two-dimensional building acquisition. We propose methods for identification and quantification of differences between independently acquired regions, and we present a systematic classification of those differences. Differences between acquired sets Rj = rij of regions rij depend on the context of observation, on the technique of observation, and so on. We distinguish topological and em geometrical differences. Topological differences refer to the interior structure of a set of regions as well as to the structure of the boundary of a single region. Geometrical differences refer to the location of the boundary of a single region or of a set of regions, independent of their representation and of the structure of the boundaries. Identification of differences requires a matching of two data sets R1 and R2 which is done here by weighted topological relationships. For the identification of topological differences between two sets R1 and R2 of regions we use the two region adjacency graphs. For an identification of geometrical differences we use the zone skeleton between two matched subsets rp1 and rq2 of the given sets. The zone skeleton is labeled with the local distances of the corresponding boundaries of the subsets; especially we investigate its density function. An example, based on two real data sets of acquired ground plans of buildings, shows the feasibility of the approach.

@inproceedings{winter*98:contributions, author = "Winter, Stephan and Ragia, Lemonia", title = "Contributions to a Quality Description of Areal Objects in Spatial Data Sets", year = "98", city = "Bonn", proceeding = "Proceedings of ISPRS Commission IV Symposium",}

Wir diskutieren zwei Kriterien zur Bewertung verschiedener Abbildungsmodelle im Rahmen der Kalibrierung einer Kamera. Die Beschreibungsläange des Datensatzes und die Stabilität/Präzision der 3D--Rekonstruktion in Abhängigkeit vom verwendeten Modell erlauben eine automatische Wahl aus einer Menge vorhandener Modelle. Am Beispiel der Off--Line Selbstkalibrierung mit verschiedenen Modellen zur Beschreibung der inneren Orientierung der Kamera demonstrieren wir diese Verfahren.

@inproceedings{abraham*97:zur, author = "Abraham, Steffen and Förstner, Wolfgang", title = "Zur automatischen Modellwahl bei der Kalibrierung von CCD-Kameras", year = "97", pages = "147-155", city = "Bonn", proceeding = "Proceedings: 19. DAGM-Symposium Mustererkennung",}

The demand for high accuracy, for example in 3D measurements or in quality control, requires high precision calibration of CCD cameras. In this paper we present a test-field-based high-precision calibration procedure with the focus on the techniques which enable an autonomous calibration: automatic control of the imaging process with geometric tests, algorithms for automatic point detection and matching, automatic determination of approximate values for the orientation parameters and automatic model selection of the interior orientation parameters. Finally, the parameters are estimated in an iterative least squares adjustment.

@inproceedings{abraham*97:towards, author = "Abraham, Steffen and Hau, Thomas", title = "Towards Autonomous High-Precision Calibration of Digital Cameras", year = "97", pages = "82-93", city = "Bonn", proceeding = "Proceedings of SPIE Annual Meeting",}

This paper describes an approach for building extraction using Digital Surface Models (DSM) as input data. The first task is the detection of areas within the DSM which describe buildings. The second task is the reconstruction of buildings for which we apply parametric and prismatic building models. The main focus is on the detection, namely on the use of height and differential geometric information in combination. Furthermore, recent results for the extraction of roof structures as first step towards the extraction of polyhedral building descriptions are presented.

@inproceedings{brunn*97:extracting, author = "Brunn, Ansgar and Weidner, Uwe", title = "Extracting Buildings from Digital Surface Models", year = "97", city = "Bonn", proceeding = "IAPRS: 3D Reconstruction and Modeling of Topographic Objects",}

The paper discusses the impact of scene and assessment models for videometry. Full automation of calibration and orientation procedures appears to be as necessary for enlarging the field of applications as the use of explicit geometric and semantic scene knowledge. The focus on achieving highest possible accuracy needs to be embedded into a broader context of scene analysis. Examples demonstrate the feasibility of tools from Computer Vision for image metrology.

@inproceedings{förstner*97:automatic, author = "Förstner, Wolfgang and Gülch, Eberhard", title = "Automatic Orientation and Recognition in Highly Structured Scenes", year = "97", city = "Bonn", proceeding = "Proceedings of SPIE Annual Meeting 1997 (to appear)",}

We propose a model-based approach to automated 3D extraction of buildings from aerial images. The semantics of the concept building is used to control and to evaluate building extraction in all stages of the process. It is encoded by means of generic 3D object modeling, which describes thematic and geometric constraints on the spatial appearences of buildings, and 2D image modeling, which integrates sensor and illumination modeling to describe the projective appearences of buildings, specific for the given aerial imagery. 3D object and 2D image modeling are tightly coupled within a multi-layered framework which reveals a is-part-of -hierachy of 3D building parts and their corresponding projective descriptions. The overall strategy follows the paradigm of hypotheses generation and verification and combines bottom-up (data-driven) and top-down (model-driven) processes. Due to the explicit representation of well defined processing states in terms of model-based 2D and 3D descriptions at all levels of modeling and data aggregation our approach reveals a great potential for a reliable building extraction.

@inproceedings{fischer*97:integration, author = "Fischer, André and Kolbe, Thomas H. and Lang, Felicitas", title = "Integration of 2D and 3D Reasoning for Building Reconstruction using a Generic Hierarchical Model", year = "97", city = "Bonn", proceeding = "Workshop on Semantic Modeling for the Acquisition of Topographic Information from Images and Maps SMATI",}

There is an increasing request for 3D data on city objects of all kinds, confirmed by a recent European wide study on 3D city models. To acquire 3D information in urban areas still is costly, only automated or at least semi-automatic methods appear feasible in the long run to reach the cost-effectiveness, necessary for a broad application. The variety of tasks and available sensor data is very large, which puts high requirements on the design of methods and the flexibility of the acquisition process. This paper discusses the requirements we have encountered so far. We present the design and current status of a semi-automatic system for 3D building acquisition. We demonstrate the potential for handling a variety of applications, using different sensor data under different initial conditions.

@inproceedings{gülch97:application, author = "Gülch, Eberhard", title = "Application of Semi-Automatic Building Acquisition", year = "97", city = "Bonn", proceeding = "Proceedings of Ascona Workshop 1997 Automatic Extraction of Man-Made Objects from Aerial and Space Images (to appear)",}

Developing a system for semiautomatic building acquisition is a complex process, that requires constant integration and updating of software modules and user interfaces. To facilitate these processes we apply an object-oriented design not only for the data but also for the software involved. We use the Unified Modeling Language (UML) to describe the object-oriented modeling of the system in different levels of detail. We can distinguish between use cases from the users point of view, that represent a sequence of actions, yielding in an observable result and the use cases for the programmers, who can use the system as a class library to integrate the acquisition modules in their own software. The structure of the system is based on the Model-View-Controller (MVC) design pattern. An example from the integration of automated texture extraction for the visualization of results demonstrate the feasibility of this approach.

@inproceedings{gülch*97:object-oriented, author = "Gülch, Eberhard and Müller, Hardo", title = "Object-oriented software design in semiautomatic building extraction", year = "97", city = "Bonn", proceeding = "Proceedings Integrating Photogrammetric Techniques with Scene Analysis and Machine Vision III",}

If spatial homogeneity is restricted to only a translation invariance then Gibbs random fields provide effective means for probabilistic modeling of homogeneous or piecewise-homogeneous scalar data on finite rectangular 2D grids. We discuss basic features of novel models with multiple pairwise interactions between the signals in grid sites. These models show good results in simulating and segmenting piecewise--homogeneous image textures. They differ from more widely known ones, such as the autobinomial or Gauss-Markov models, in that both the interaction structure and strengths are learnt from a given training sample. A new learning approach, based on conditional maximum likelihood estimates of the model parameters, provided that the training sample may rank a feasible top place in a parent population, is proposed. We applied the model to reproduce geomorphometric patterns derived from a terrain classification method. The study is aimed at showing that texture analysis, based on a quantification of neighborhood relationships, can be used to descriminate landform types. The texture segmentation shows good correlations with the terrain classification used for learning the Gibbs model parameters. The approach could be valuable in quantifying geomorphometric structures and manual terrain classification schemes.

A semiautomatic building extraction system has been extended by an automatic matching tool. It is used for an automatic measurement of building-heights and a semiautomatic determination of ground-heights. The object-oriented design of this matching tool gives the motivation for a design pattern of a general matching tool. This design pattern describes the object-oriented design of implementing several matching techniques within one system. It is able to be applied to several kinds of topographic objects, if a matching technique is known. We show as an example the implementation of a point matching tool, which uses the matching techniques Intensity Correlation and Gradient Correlation as a refinement of Feature Vector Matching.

@inproceedings{müller97:designing, author = "Müller, Hardo", title = "Designing an object-oriented matching tool", year = "97", city = "Bonn", proceeding = "IAPRS: 3D Reconstruction and Modeling of Topographic Objects",}

Gebäudedaten bilden einen wesentlichen Bestandteil der in den letzten Jahren mehr und mehr nachgefragten dreidimensionalen Daten digitaler Stadtmodelle. Solche Daten sind häufig nicht verfügbar oder genügen nicht den Anforderungen der jeweiligen Anwendung. Aus diesen Gründen sind oftmals eine Ersterfassung und Fortführung notwendig. Die Lösung beider Aufgaben ist mit den heute eingesetzten Verfahren i.a. mit einem hohen Aufwand verbunden, so daß eine Automation und hierdurch eine Minimierung des Aufwandes wünschenswert sind. Thema dieser Arbeit ist die Gebäudeerfassung aus Digitalen Oberflächenmodellen (DOM). Diese repräsentieren nicht nur Information über die topographische Oberfläche, wie Digitale Höhenmodelle (DHM), sondern auch Information über die in einem Gebiet befindliche Bebauung und andere aus der topographischen Oberfläche herausragende Objekte. Es wird ein Verfahren zur automatischen Gebäudeerfassung aus DOM vorgestellt und die Eignung von DOM als Datengrundlage für die Gebäudeerfassung untersucht. Die alleinige Nutzung von DOM zur Gebäudeerfassung unter Verwendung parametrischer und prismatischer Gebäudemodelle stellt einen neuen Ansatz dar und ermöglicht den Einsatz von Sensoren, die bislang nicht zum Zweck der Gebäudeerfassung verwendet wurden, \zB luftgestützte Laserabtaster. Die in der Literatur bekannten Verfahren zur Gebäudeerfassung setzen DOM oft ausschließlich zum Zweck der Gebäudedetektion ein. Das in ihrer Auswertung liegende Potential auch zur Gebäuderekonstruktion wird nur wenig ausgeschöpft. Als zusätzliche Datenquelle werden Geoinformationssysteme (GIS) genutzt, die i.a. nur zweidimensionale Information enthalten. Die verwendeten Objekt-, Sensor- und Datenmodelle werden vorgestellt und begründet, sowie die verfolgte Auswertestrategie und die Auswertemethoden dargestellt. Auswertestrategie und Objektmodelle besitzen Allgemeingültigkeit und werden auch in der einen oder anderen Form in anderen Verfahren zur Gebäudeerfassung genutzt. Die aus der Diskussion dieser Verfahren abgeleiteten Schlußfolgerungen finden ihren Niederschlag in dem entwickelten Verfahren. Das neue Verfahren umfaßt die Detektion und Rekonstruktion von Gebäuden auf der alleinigen Datengrundlage von DOM und -- falls verfügbar -- zusätzlichen GIS-Daten, sowie eine Änderungsdetektion zum Zwecke der GIS-Fortführung. Ein wesentlicher Bestandteil der Arbeit ist die qualitative und quantitative, auf der Grundlage von abgeleiteten und diskutierten Kenngrößen durchgeführte Bewertung der erzielten Ergebnisse. Ferner werden die Ergebnisse anhand von Referenzdaten mit denen anderer Verfahren verglichen. Die erzielten Ergebnisse zeigen die prinzipielle Eignung von DOM zur Gebäudeerfassung mit dem hier entwickelten Verfahren an. Einschränkungen hinsichtlich der alleinigen Nutzung von DOM und Probleme hinsichtlich der Modellierung werden diskutiert und Lösungsansätze aufgezeigt. Abstract During the last years the need for three-dimensional data describing urban areas increased. This data include building data, which in many cases are not available or do not fulfill the requirements of a special application. Due to these reasons, the capture and update of this data are necessary. Available techniques to solve these tasks are often expensive. Therefore, automization and thereby increasing the efficiency of data acquisition are required. The topic of the thesis is building extraction from Digital Surface Models (DSM). Such surface models do not only contain information about the topographic surface like Digital Elevation Models (DEM), but also about buildings and other objects higher than the topographic surface. An approach to automatic building extraction from DSM is presented and the suitability of DSM as input data for this purpose is investigated. The use of DSM for building extraction using parametric and prismatic building models as only input data is a new approach. It offers the possibility to use sensors like airborne laser scanners, which have not been used for the purpose of building data capture up to now. Previously published approaches dealing with building extraction often use DSM for building detection only. The inherent potential of DSM with respect to building reconstruction was explored only by a few authors, but either restricted to simple building models or by using digital images as additional and necessary input data. Besides DSM, GIS data is used in this approach as additional data source, which mostly contain only two-dimensional information. The object, sensor, and data models, which are used in this thesis, are discussed and motivated, and the strategy and methods for data exploitation are presented. The strategy and the object models - possibly in a modified form - are also used in other approaches to building extraction. Some of these approaches are discussed and the conclusions are reflected in the presented new approach. The new approach covers the detection and reconstruction of buildings using DSM as only input data. GIS data - if available - can be incorporated, allowing also change detection for the purpose of updating the GIS. The results of the different steps are evaluated qualitatively and quantitatively. The quality measures are derived and discussed. Furthermore, the results are compared to those of other approaches using reference data. The achieved results show that DSM are in principle suitable for building extraction using the described approach. Restrictions with respect to the only use of DSM and problems concerning the modeling are discussed and solutions are proposed.

This paper describes an approach to building extraction using Digital Surface Models (DSM) as input data. The approach consists of building detection and reconstruction using parametric and prismatic building models. The main focus is on the extraction of roof structures, an extension of the previously published work, as first step towards the extraction of polyhedral building descriptions in order to also allow the extraction of complex buildings.

@inproceedings{weidner97:digital, author = "Weidner, Uwe", title = "Digital Surface Models for Building Extraction", year = "97", city = "Bonn", proceeding = "Automatic Extraction of Man-Made Objects from Aerial and Space Images (II)",}

This paper deals with the derivation of a symbolic surface description exploiting the information of multiple images while using a minimum of domain knowledge. We present a new concept for segmenting surfaces by fusing multiple images both on the iconic and on the symbolic description level. In a first step a local 3D-reconstruction and interpretation is derived based on the result of a polymorphic feature extraction. It serves as prior information for a second step which refines the initial segmentation using the radiometric image content. Examples of the proposed procedure are presented for the application of 3D-building reconstruction from aerial images.

@inproceedings{brunn*96:procedure, author = "Brunn, Ansgar and Lang, Felicitas and Förstner, Wolfgang", title = "A Procedure for Segmenting Surfaces by Symbolic and Iconic Image Fusion", year = "96", pages = "11-20", city = "Bonn", proceeding = "Mustererkennung 96, Proceeding of the DAGM 96",}

In various projects we investigate on the extraction of buildings on different type and representation of data. This paper presents a strategy for 3D building acquisition which combines different approaches based on different levels of description. The approach consists of detection of regions of interest and automatic and semiautomatic reconstruction of object parts and complete buildings. We incorporate the approach in a global concept of interaction between scene and sensors for image interpretation.

@inproceedings{brunn*96:multi-layer, author = "Brunn, Ansgar and Lang, Felicitas and Gülch, Eberhard and Förstner, Wolfgang", title = "A Multi-Layer Strategy for 3D Building Acquisition", year = "96", city = "Bonn", proceeding = "Proceeding of IAPR-TC7 Workshop",}

Der Beitrag stellt photogrammetrische Systeme zur automatischen Vermessung von Oberflächen vor, die seit einigen Jahren im praktischen Einsatz sind, und präsentiert aktuelle Forschungs-und Entwicklungsarbeiten zur Gebäudeextraktion.

@inproceedings{förstner96:automatische, author = "Förstner, Wolfgang", title = "Automatische 3D-Objekterfassung und -erkennung", year = "96", city = "Bonn", proceeding = "gekürzte Fassung des Vortrags auf der Ringvorlesung Bildverarbeitung und Mustererkennung an der Universität Bonn im WS 1995/96",}

The paper discusses objections against performance characterization of vision algorithms and explains their motivation. Short and long term arguments are given which overcome these objections. The methodology for performance characterization is sketched to demonstrate the feasibility of empirical testing of vision algorithms.

@inproceedings{förstner96:10, author = "Förstner, Wolfgang", title = "10 Pros and Cons Against Performance Characterization of Vision Algorithms", year = "96", city = "Bonn", proceeding = "Workshop on "Performance Characteristics of Vision Algorithms"",}

There is an increasing request for 3D data on city objects of all kinds for urban design, confirmed by a recent European wide study on 3D city models. To acquire 3D information in urban areas still is costly, only automated or at least semi-automatic methods appear feasible in the long run to reach the cost-effectiveness, necessary for a broad application. This paper presents a semi-automatic system for 3D building acquisition from various sensor data, mainly, however, from stereo pairs of digitized aerial images. The operator is supported by various automated modules. Very complex buildings can be modeled by a combination of volumetric primitives. The system does not require stereo-viewing and is such suitable also for non-photogrammetrists. The output of the process are 3D volumetric primitives, ready for further analysis in CAD systems or for visualization purposes in combination with automatically extracted texture. We present results from the acquisition of 3D building information in a suburban area and the centers of two cities and give more details on the acquisition times and the quality of the derived data.

@inproceedings{gülch96:extraction, author = "Gülch, Eberhard", title = "Extraction of 3D objects from aerial photographs", year = "96", city = "Bonn", proceeding = "Proceedings COST UCE Action 4 Workshop "Information Systems and Processes for urban civil engineering"",}

An easy usable system for the semi-automatic acquisition of detailed 3D building descriptions from a multitude of images is provided. This approach tackles robustly and efficiently most of the problems of 3D building reconstruction, namely occlusions, inverse mapping, and noise. The 3D modeling is based on Constructive Solid Geometry (CSG) and various automated and supporting tools. Our experiences on the acquisition of an extensive scene are evaluated. Zusammenfassung Monokulares Stereo System für die Erfassung komplexer 3D-Gebäudebeschreibungen. Für die semiautomatische Erfassung detaillierter 3D-Gebäudebeschreibungen aus einer Menge von Bildern wird ein einfach handhabbares System vorgestellt. Auf robuste und effiziente Weise werden die meisten Probleme bei der 3D-Gebäuderekonstruktion -- Verdeckungen, inverse Abbildung und Rauschen -- angegangen. Die 3D-Modellierung basiert auf der Constructive Solid Geometry (CSG) und zahlreichen automatisierten und unterstützenden Werkzeugen. Unsere Erfahrungen mit einer grossflächigen Erfassung werden evaluiert.

@inproceedings{gülch96:one-eye, author = "Gülch, Eberhard", title = "One-Eye Stereo System for the Acquisition of Complex 3D Building Descriptions", year = "96", city = "Bonn", proceeding = "GIS",}

A Markov random field model with a Gibbs probability distribution (GPD) is proposed for describing particular classes of grayscale images which can be called spatially uniform stochastic textures. The model takes into account only multiple short- and long-range pairwise interactions between the gray levels in the pixels. An effective learning scheme is introduced to recover a structure and strength of the interactions using maximal likelihood estimates of the potentials in the GPD as desired parameters. The scheme is based on an analytic initial approximation of the estimates and their subsequent refinement by a stochastic approximation. Experiments in modelling natural textures show the utility of the proposed model.

@inproceedings{gimelfarb96:texture, author = "Gimelfarb, Georgy L.", title = "Texture Modeling by Multiple Pairwise Pixel Interactions", year = "96", city = "Bonn", proceeding = "IEEE T-PAMI",}

The bottleneck of today's automation of image orientation is the identification of control points for exterior orientation. A solution for this problem is presented. It is based on 3D-wireframe models of buildings as ground control points. The paper describes the setup of a database of such control points.

@inproceedings{läbe*96:3d-wireframe, author = "Läbe, Thomas and Ellenbeck, Karl Heiko ", title = "3D-Wireframe Models As Ground Control Points For The Automatic Exterior Orientation", year = "96", city = "Bonn", proceeding = "Proceedings of 18th ISPRS Congress",}

3D-city information is crucial for a number of applications in city planning, environmental control or for telecommunication. We describe a semiautomatic system for acquiring the 3D-shape of buildings as topographic objects. Buildings are either modeled as a freely structured union of basic shape primitives or as prisms with an arbitrary ground plan, covering a large percentage of existing buildings. Interaction takes place in only one image, requiring the operator to specify the approximate structure and shape of the buildings. 3D-reconstruction including both, height determination and form adaptation, is performed automatically using various matching tools. The paper describes the features of the system and reports on its efficiency based on an extensive test.

@inproceedings{lang*96:3d-city, author = "Lang, Felicitas and Förstner, Wolfgang", title = "3D-City Modeling with a Digital One-Eye Stereo System", year = "96", city = "Bonn", proceeding = "Proc. ISPRS Congress Comm. IV",}

This paper presents a new concept for 3D-surface reconstruction, which infers domain specific local 3D-structures in space from its observed local 2D-structures in multiple images using polymorphic relational image descriptions. A 3D-aggregation can combine these local 3D-structures and thus results in a 3D-boundary representation of man-made objects being useful for different analyses and simulations.

@ARTICLE{lang*96:surface, author = "Lang, Felicitas and Förstner, Wolfgang", title = "Surface Reconstruction of Man-Made Objects using Polymorphic Mid-Level Features and Generic Scene Knowledge", year = "1996", pages = "193-202", city = "Bonn",}

We propose to use Constraint Logic Programming (CLP) for the specification and implementation of Qualitative Reasoning (QR) problemsthat are specialized Constraint Satisfaction Problems. Although it has long been recognized that many frameworks, like the QSIM algorithm \cite{kuipers94:qualitative} can be viewed as a set of constraint satisfaction problems (CSP), the area of QR is relatively unknown in the CLP literature. In this paper we would like to present, through the language of QSIM, one set of questions and problems analyzed in the area of QR and to present the CLP specification of the key algorithm of QSIM, the c-filter. We show how the basic constraints of QSIM are specified and describe the technical aspects of our implementation.

@inproceedings{teleki96:embedding, author = "Teleki, László", title = "Embedding Qualitative Reasoning into Constraint Logic Programming", year = "96", city = "Bonn", proceeding = "Proceedings of the Workshop on Constraint Programming Applications, in conjunction with the Second International Conference on Principles and Practice of Constraint Programming (CP96)",}

We propose to use Constraint Logic Programming (CLP) for the specification and implementation of Qualitative Reasoning (QR) problems that are specialized Constraint Satisfaction Problems. The use of CLP has two advantages: (i) CLP gives a well defined and understood logical framework for the problem specification, and (ii) CLP is not only a logical framework, it is also a family of languages specially developed for solving classes of CSP problems. Thus we obtain a class of powerful implementation languages for rapid prototyping. To illustrate the steps of specification and implementation we describe in detail the core of the QSIM algorithm \cite{kuipers94:qualitative}, namely the filtering of the state transitions in the CLP framework. We show how the basic constraints are specified in this framework and describe the technical aspects of an implementation. We want to demonstrate the advantages of CLP through an example for a large and complex qualitative reasoning algorithm.

@inproceedings{teleki96:constraint, author = "Teleki, László", title = "Constraint Logic Programing - a Framework for Qualitative Reasoning", year = "96", city = "Bonn", proceeding = "Qualitative Reasoning, The Tenth International Workshop",}

Motivated by the test data sets of ISPRS WG III/3 on image understanding we investigate the feasibility of building extraction using high-resolution Digital Surface Models (DSM) as input data, which do not only contain information about the topographic surface like Digital Elevation Models (DEM), but also information about the buildings. The steps of the proposed procedure increasingly use explicit domain knowledge, specifically geometric constraints in the form of parametric and prismatic building models. The reconstruction of the prismatic models and the selection of the models are based on the principle of Minimum Description Length (MDL). In Weidner/Förstner (1995): Towards Automatic Building Reconstruction from High Resolution Digital Elevation Models we already described the general strategy of our approach, including building detection and building reconstruction. The main extensions of this contribution consists of the automatic selection of the model for the reconstruction of buildings. This selection is based on the principle of Minimum Description Length (MDL), as well as the reconstruction of prismatic building models. In addition, we also discuss the possible use of information from GIS or maps in our approach.

@inproceedings{weidner96:approach, author = "Weidner, Uwe", title = "An Approach to Building Extraction from Digital Surface Models", year = "96", city = "Bonn", proceeding = "Proceedings of 18th ISPRS Congress",}

Dieses Papier ist die Zusammenfassung eines Vortrages während der wissenschaftlichen Jahrestagung der DGPF 1995 und beschreibt einen MDL-basierten Ansatz zur Rekonstruktion von Polygonen unter Nutzung lokaler und globaler Restriktionen.

@inproceedings{weidner96:mdl-basierte, author = "Weidner, Uwe", title = "MDL-basierte Formrekonstruktion zur Gebäudeextraktion", year = "96", city = "Bonn", proceeding = "Wissenschaftliche Jahrestagung der DGPF 1995",}

In diesem Beitrag wird eine Methode zur Bestimmung der topologischen Relationen zwischen zwei in ihrer Lage unsicheren Regionen vorgestellt. Die Bestimmung basiert auf den Extremwerten einer Abstandsfunktion entlang eines Skeletts durch unsichere Schnittmengen zwischen den Regionen. Mit einer solchen Reprasentation vertiefen wir die Relationskonzepte der 9-Intersektion, indem wir über topologisch invariante Merkmale hinaus auch metrische Information erhalten, um sie zur Unsicherheit in Bezug zu setzen. Die Beobachtung der Abstände wird hier statistisch modelliert. Mit einer Bayes-Klassifikation erhalten wir Abstandsklassen, aus denen wir auf die topologische Relation zurückschließen und deren Wahrscheinlichkeit wir so angeben können.

@inproceedings{winter96:beobachtungsunsicherheit, author = "Winter, Stephan", title = "Beobachtungsunsicherheit und topologische Relationen", year = "96", city = "Bonn", proceeding = "1996",}

Considering uncertainty of spatial data in any GIS analysis is a theme of actual research, but far from practicability. One prerequisite is the availability of descriptions concerning uncertainty (meta-data), and another is the attachment of uncertainty to spatial analysis. In our contribution we propose a method to determine the topological relation between two positional uncertain regions. The decision is based on morphological distances along a skeleton through uncertain intersection sets. These measures are equivalent to the known representation by intersection sets, but yield additionally metric information. These distances are applied to a statistical model of the observation process. A Bayesian classification yields distance classes which allow to deduce the topological relationship. We discuss also importance and perspectives of the method.

@inproceedings{winter96:distances, author = "Winter, Stephan", title = "Distances for Uncertain Topological Relations", year = "96", city = "Bonn", proceeding = "ESF-NSF Summer Institute in Geographic Information",}

@inproceedings{winter*96:report, author = "Winter, Stephan and Car, Adrijana", title = "Report from ESF/NCGIA Summer Institute in GIS, Berlin", year = "96", city = "Bonn", proceeding = "1996",}

The paper discusses the modeling necessary for recovering man made objects - in this case buildings - in complex scenes from digital imagery. The approach addresses all levels of image analysis for deriving semantically meaningful descriptions of the scene from the image, via the geometrical/physical model of the objects and their counterparts in the image. The central link between raster image and scene are network-like organized aspects of parts of the objects. This is achieved by generically modelling the objects using parametrized volume primitives together with the application specific constraints, which seems to be adequate for many types of buildings. The paper sketches the various interrelationships between the different models and their use for feature extraction, hypothesis generation and verification.

@inproceedings{braun*95:models, author = "Braun, Carola and Kolbe, Thomas H. and Lang, Felicitas and Schickler, Wolfgang and Steinhage, Volker and Cremers, Armin B. and Förstner, Wolfgang and Plümer, Lutz", title = "Models for Photogrammetric Building Reconstruction", year = "95", pages = "109-118", city = "Bonn", proceeding = "Computer & Graphics, Vol. 19",}

Mid-level processes in vision are understood to produce structured descriptions of images without relying on very specific semantic scene knowledge. Automatic building extraction can use geometric models to a large extent. Geometric hypotheses may be inferred from the given data in 2D or 3D and represent elementary constraints as incidence or collinearity or more specific relations as symmetries. The inferred hypothesis may lead to difficulties during spatial inference due to noise and to inconsistent and mutually dependent constraints. The paper discusses the selection of mutually not-contradicting constraints via robust estimation and the selection of a set of independent constraints as a prerequisite for an optimal estimation of the objects shape. Examples from the analysis of image and range data are given.

@inproceedings{förstner95:mid-level, author = "Förstner, Wolfgang", title = "Mid-Level Vision Processes for Automatic Building Extraction", year = "95", pages = "179-188", city = "Bonn", proceeding = "Automatic Extraction of Man-Made Objects from Aerial and Space Images",}

The paper discusses the problem areas in establishing 3D-Geoinformation Systems. Among the many applications it restricts to the 3D-modelling of cities. Acquisition methods for 3D-data and examples for the use of such 3D-models are presented. Finaly, a 2D-representation for 3D-objects with vertical walls but without passages is proposed which may be used for storing buildings and which may form a link to CAD-systems.

@inproceedings{förstner95:gis, author = "Förstner, Wolfgang", title = "GIS - The Third Dimension", year = "95", city = "Bonn", proceeding = "Proc. IUSM WG on GIS/LIS Workshop "Current Status and Challenges of Geoinformation Systems",",}

@INPROCEEDINGS{förstner95:matching, author = {Wolfgang Förstner}, title = {A Unified Framework for the Automatic Matching of Points and Lines in Multiple Oriented Images}, year = {1995}, pages = {173-183}}

The paper discusses tools of diagnostics and robustness in the context of automating vision. Motivation is the building of so-called traffic light programs which contain a reliable selfdiagnosis enabling to chain vision modules. Special attention is payed to show the prerequisites for using tools for quality evaluation. The paper concludes with open questions.

@inproceedings{förstner95:role, author = "Förstner, Wolfgang", title = "The Role of Robustness in Computer Vision", year = "95", city = "Bonn", proceeding = "Proc. Workshop "Milestones in Computer Vision"",}

@inproceedings{förstner*95:proc., author = "Förstner, Wolfgang and Ditze, G.", title = "Proc. of the Second Course in Digital Photogrammetry", year = "95", city = "Bonn", proceeding = "1995",}

The paper deals with an approach for extracting the 3D-shape of buildings from high resolution Digital Elevation Models (DEMs), having a grid resolution between 0.5 and 5m. The steps of the proposed procedure increasingly use explicit domain knowledge, specifically geometric constraints in the form of parametric and prismatic building models. A new MDL-based approach generating a polygonal ground plan from segment boundaries is given. The used knowledge is object related making adaption to data of different density and resolution simple and transparent.

@inproceedings{förstner*95:towards, author = "Förstner, Wolfgang and Weidner, Uwe", title = "Towards Automatic Building Reconstruction from High Resolution Digital Elevation Models", year = "95", pages = "38-49", city = "Bonn", proceeding = "ISPRS Journal",}

The paper presents a new approach for the reconstruction of polygons using local and global conctraints. The MDL-based solution is shown to be useful for analysing range and image data of buildings. Paper at 17th DAGM symposium '95, Bielefeld, September 13.-15. 1995

@inproceedings{förstner*95:model-based, author = "Förstner, Wolfgang and Weidner, Uwe and Brunn, Ansgar", title = "Model-based 2D-Shape Recovery", year = "95", pages = "260-268", city = "Bonn", proceeding = "Mustererkennung",}

The paper describes a new approach to image segmentation. It accepts the inherent deficiencies occuring when extracting low-level features and when dealing with the complexity of real scenes. Image segmentation therefore is understood as deriving a rich symbolic description useful for tasks such as stereo or object recognition in outdoor scenes. The approach is based on a polymorphic scheme for simultaneously extracting points, lines and segments in a topologically consistent manner, together with their mutual relations derived from the feature adjacency graph (FAG) thereby performing several grouping steps which gradually use more and more specific domain knowledge to achieve an optimal image description. The heart of the approach is 1.) a detailed analysis of the FAG and 2.) a robust estimation for validating the found geometric hypotheses. The analysis of the FAG, derived from the exoskeleton of the features, allows to detect inconsistencies of the extracted features with the ideal image model, a cell-complex. The FAG is used for finding hypotheses about incidence relations and geometric hypotheses, such as collinearity or parallelity, also between non-neighbored points and lines. The M-type robust estimation is used for simultaneously eliminating wrong hypotheses on geometric relationships. It uses a new argument for the weighting function.

@inproceedings{fuchs*95:polymorphic, author = "Fuchs, Claudia and Förstner, Wolfgang", title = "Polymorphic Grouping for Image Segmentation", year = "1995", city = "Cambridge, U.S.A.", proceeding = "5th ICCV" }

One of the central tasks in Photogrammetry and Computer Vision is the localization and reconstruction of objects in the scene. Localization aims at determining the pose , i. e. the position and the orientation of an object. It assumes the form of the object to be known or at least to be known up to some structural or numerical parameters and the mutual relation between the reference frames, i. e. coordinate systems of the object and the cameras to be determined. Reconstruction , on the other hand aims at determining the form possibly also the structure of the object. The form description need not, but may be related to a e. g. object centred, reference coordinate system. In all cases the central tasks is to match the description of one or several images to the description of the images or objects, i.e. to establish correspondence. In all cases automating localization and reconstruction requires to establish the correspondence or match between several images or between one or several images and a model. We therefore may distinguish several cases: * Image Matching * Object Localization * Object Reconstruction which are discussed in this publication.

@inproceedings{lang*95:matching, author = "Lang, Felicitas and Förstner, Wolfgang", title = "Matching Techniques", year = "95", city = "Bonn", proceeding = "Proc.: 2nd Course in Digital Photogrammetry",}

We present a semi-automatic system for building extraction that has been developed at the Institute for Photogrammetry at the University of Bonn. Digitized images are used as data source. Data capture takes place at a simple workstation. Single point measurement is replaced by the measurement of building models. Automation supports the operator and thus increases the system performance. The representation of building models allows a link to CAD, GIS and computer graphics.

@inproceedings{lang*95:one-eye, author = "Lang, Felicitas and Löcherbach, Thomas and Schickler, Wolfgang", title = "A one-eye Stereo System for Semi-Automatic 3D-Building Extraction", year = "95", city = "Bonn", proceeding = "Geomatics Info Magazine",}

The paper is an abstract of a talk concerning topics related to the interpretation of Digital Elevation Models.

@inproceedings{weidner95:interpretation, author = "Weidner, Uwe", title = "Interpretation von Digitalen Höhenmodellen", year = "95", city = "Bonn", proceeding = "Talk at SFB Workshop Das Relief",}

@inproceedings{winter95:dt, author = "Winter, Stephan", title = "DT - Modul zur Distanztransformation", year = "95", city = "Bonn", proceeding = "1995",}

@inproceedings{winter95:fourth, author = "Winter, Stephan", title = "Fourth Symposium on Large Spatial Databases (SSD 95)", year = "95", city = "Bonn", proceeding = "1995",}

Topological reasoning is important for speeding up spatial queries, e.g. in GIS or in AI (robotics). While topological relations between spatial objects in the vector model (R2) are investigated thoroughly, we run into inconsistencies in the raster model (Z2). But instead of reducing our requirements in case of reasoning in raster images we change from simple raster to a cellular decomposition of R2 - what we call a hyper-raster - which is also discrete, but preserves the topology of R2. The finite representation limits the computational effort against the vector model. We will introduce a data structure for the hyper-raster, which represents regions, curves and points. Then we will present algorithms for digitization and elementary image processing in hyper-raster. With that the intersection sets, as needed for the determination of a topological relation between two objects, are calculated simply by logical joins of binary images. Without extending our model we can also compute further refinements of the relationship. We will show that applying the 9-intersection, originally developed for the vector model, to the hyper-raster leads exactly to the set of known eight relations between regions without holes in R2.

@inproceedings{winter95:topological, author = "Winter, Stephan", title = "Topological Relations between Discrete Regions", year = "95", pages = "310-327", city = "Bonn", proceeding = "Advances in Spatial Databases Proceedings of the Fourth International Symposium on Large Spatial Databases (SSD 95)",}

The paper discusses the modeling necessary for recovering man made object - in this case buildings - in complex scenes from digital imagery. The approach addresses all levels of image analysis for deriving semantically meaningful descriptions of the scene from the image, via the geometrical / physical model of the objects and their counterparts in the image. The central link between raster image and scene are network-like organized aspects of parts of the objects. This is achieved by generically modelling the objects using parametrized volume primitives together with the application specific constraints, which seems to be adequate for many types of buildings. The paper sketches the various interrelationships between the different models and their use for feature extraction, hypothesis generation and verification.

@inproceedings{braun*94:models, author = "Braun, Carola and Kolbe, Thomas H. and Lang, Felicitas and Schickler, Wolfgang and Steinhage, Volker and Cremers, Armin B. and Förstner, Wolfgang and Plümer, Lutz", title = "Models for Photogrammetric Building Reconstruction", year = "94", city = "Bonn", proceeding = "1994",}

The paper presents a framework for extracting low level features, namely points, edges and segments from digital images. It is based on generic models for the scene, the sensing and the image. Its main goal is to explicitely exploit the information content of the image as far as possible. This leads to new techniques for deriving image parameters, to either the elimination or the elucidation of "buttons", like thresholds, and to interpretable quality measures for the results, which may be used in subsequent steps. The feature extraction is based on local statistics of the image function, namely the average squared gradient and on the regularity of the image function with respect to junctions and circular symmetric features as special cases of Bigun's (1990) spiral type features. Methods are provided for blind estimation of a signal dependent noise variance, for feature preserving restoration, for feature detection and classification and for the precise location of general edges and points. Their favorable scale space properties are discussed. In alls steps thesholding and classification is based on proper test statistics reducing threshold selection to choosing a significance level.

@INPROCEEDINGS{förstner94:framework, author = {Förstner, Wolfgang}, title = {A Framework for Low-Level Feature Extraction}, booktitle = {ECCV'94}, year = {1994}, volume = {801/1994}, pages = {383-394}, doi = {10.1007/BFb0028370} }

Increasing the performance of Computer Vision algorithms requires both, robust procedures for handling non-modelled errors and diagnostic tools for achieving autonomy in the evaluation of the achieved results. The role of diagnostic tools for model evaluation and performance prediction is discussed. Quality or performance refers to: 1. the precision of the estimated quantities (efficiency) 2. the sensitivity of the estimated quantities with respect to systematic and gross errors. 3. the design of the experiment or the used actual data. 4. the correctness of results and of reports on the correctness of the result. The performance may be evaluated a. by controlled tests using simulated or real data. This is necessary to prove either the usefulness of the algorithms or the adequateness of the used model. b. by diagnostic tools. This is necessary for achieving autonomy in the chain of automatic procedures within a complete system where generally no reference data are available. The performance of Computer Vision algorithms can be significantly increased by diagnostic tools, both by detecting singular or weak configurations within high break down estimation, e. g. RANSAC, and by providing a highly reliable selfdiagnosis of the algorithm itself using the internally available redundancy. Results from extensive empirical tests demonstrate the feasibility of the proposed tools.

@inproceedings{förstner94:diagnostics, author = "Förstner, Wolfgang", title = "Diagnostics and Performance Evaluation in Computer Vision", year = "94", pages = "11-25", city = "Bonn", proceeding = "Performance versus Methodology in Computer Vision, NSF/ARPA Workshop",}

The paper presents a concept for analysing the quality of relational descriptions of digital images. The investigations are based on the relational description automatically derived by a new coherent procedure for feature extraction providing a feature adjacency graph containing points, edges and segments and their relations. A new notion of scale (integration scale) is introduced, relating to a non linear function of the image, providing new stable descriptions. Based on the feature extraction we analysed the quality of the relational descriptions in dependency on the signal-to-noise ratio and on the control parameters of the feature extraction process, i.~e. the significance level, the smoothing scale and the integration scale. First results on the quality of the features, focussing on their existence, distinct attributes and relations are given. The scope of this research is to predict the quality, especially probabilities of components of the relational description from a few measures depending on noise, scale and local properties of the image content. This is motivated by the applications we are dealing with, namely extracting man-made 2D or 3D structures by grouping procedures or image matching as both tasks are optimization problems where the probability of the unknown 2D or 3D model has to be maximized.

@inproceedings{förstner*94:noise, author = "Förstner, Wolfgang and Lang, Felicitas and Fuchs, Claudia", title = "On the Noise and Scale Behaviour of Relational Descriptions", year = "94", city = "Bonn", proceeding = "1994",}

Textures in images have a natural order, both in orientation and multiple narrow-band frequency, which requires to employ multichannel local spatial/frequency filtering and orientation selectivity, and to have a multiscale characteristic. Each channel covers one part of whole frequency domain, which indicates different information for the different texton. Gabor filter, as a near orthogonal wavelet used in this paper, has orientation selectivity, multiscale property, linear phase and good localization both in spatial and frequency domains, which are suitable for texture analysis. Gabor filters are employed for clustering the similarity of same type of textons. Gaussian filters are also used for detection of normal image edges. Then hybrid texture and nontexture gradient measurement is based on fusion of the difference of amplitude of the filter responses between Gabor and Gaussian filters at neighboring pixels by mainly using average squared gradient. Normalization, beased on the noise response and based on maximum response are computed.

@inproceedings{förstner*94:gabor, author = "Förstner, Wolfgang and Shao, Juliang", title = "Gabor Wavelets for Texture Edge Extraction", year = "94", city = "Bonn", proceeding = "ISPRS Commission III Symposium on Spatial Information from Digital Photogrammetry and Computer Vision",}

The paper describes an approach to reconstructing agricultural land-use areas from remotely sensed images using digital polygon maps as prior information. Our goal is to update the geometry and class label of agricultural parcels. The approach integrates an estimation of the vector polygons and the derivation of object (i. i. land-use unit) related features for the determination of the object classes. Both a feature edge model for the transition between two land-use units and the assumption of homogeneity are used to reconstruct the land-use units in a least squares approach. The theoretical concept and its technical realization are described, first results are presented and a critical evaluation of the results with a discussion of possible extension is given.

@inproceedings{löcherbach94:fusion, author = "Löcherbach, Thomas", title = "Fusion of Multi-Sensor Images and digital Map Data for the Reconstruction and Interpretation of Agricultural Land-use Units", year = "94", city = "Bonn", proceeding = "Submitted to ISPRS Commission III Symposium on Spatial Information from Digital Photogrammetry and Computer Vision",}

@ARTICLE{pan94:two-level, author = "Pan, He-Ping", title = "Two-Level Global Optimization for Image Segmentation", year = "1994", pages = "21-32", city = "Bonn",}

In this paper we present an algorithm for parameterfree information-preserving surface restoration. The algorithm is designed for 2.5D and 3D surfaces. The basic idea is to extract noise and signal properties of the data simultaneously by variance-component estimation and use this information for filtering. The variance-component estimation delivers information on how to weigh the influence of the data dependent term and the stabilizing term in regularization techniques, and therefore no parameter which controls this relation has to be set by the user.

The paper is the original contribution for ECCV'94 (7 pages) as it is published by Springer. Also available are:

• Extended version (1.5 MB)
• Presented poster (19.8 MB), with additional examples

@inproceedings{weidner94:parameterfree, author = "Weidner, Uwe", title = "Parameterfree Information-Preserving Surface Restoration", year = "94", city = "Bonn", proceeding = "Computer Vision ECCV",}

Pre-processing such as filtering data in order to remove or at least reduce noise is a crucial step because information which is lost during this filtering cannot be recovered in subsequent steps. It is a well-known fact, that linear filtering does not only reduce noise, but may also lead to a loss of information due to the global smoothing, regardless of structures in the data. In order to overcome these drawbacks, we propose to use an algorithm for parameterfree information-preserving surface restoration. As we do not want to evaluate the results of the filtering only qualitatively by visual inspection, we examine the influence of pre-processing on feature extraction for digital elevation models and discuss quantities for the evaluation of these influences.

@inproceedings{winter94:information-preserving, author = "Winter, Stephan", title = "Information-Preserving Surface Restoration ...", year = "94", city = "Bonn", proceeding = "Proceedings of ISPRS Comm. III Symposium on Spatial Information from Digital Photogrammetry and Computer Vision",}

We present a concept for representing uncertain topological relations and their derivation from uncertain sets, useful for spatial and temporal reasoning in GIS. The concept is based on the notion of a stochastic boundary of a geometric set and on tests performed to decide the validity of relations between the sets. It uses the power-function to derive the probabilities of the found relations. The concept is applicable to all questions where uncertain geometric queries or analysis have to be performed.

@inproceedings{winter94:uncertainty, author = "Winter, Stephan", title = "Uncertainty in Topological Relationships in GIS", year = "94", city = "Bonn", proceeding = "Proceedings of ISPRS Comm. III Symposium on Spatial Information from Digital Photogrammetry and Computer Vision",}

@inproceedings{förstner93:image, author = "Förstner, Wolfgang", title = "Image Matching", year = "93", pages = "289-379", city = "Bonn", proceeding = "Computer and Robot Vision",}

The paper presents a concept for the model based threedimensional acquisition of buildings from single and multiple images. For solving the interpretation problem an automatic building extraction procedure requires an apropriate object model. We use parametrized models implicitely representing inherent geometric and topological contraints. The semiautomatic procedure supports the operator during the mensuration process and the final fit of the models to the image data. The implementation and first results are reported.

@ARTICLE{lang*93:semiautomatische, author = "Lang, Felicitas and Schickler, Wolfgang", title = "Semiautomatische 3D Gebaeudeerfassung aus digitalen Bildern", year = "1993", pages = "193-200", city = "Bonn",}

This paper discusses an automatic procedure for color edge extraction. It contains a procedure for robustly estimating the signal dependent components ofthe noise which is assumed to be influenced mainly by the Poisson statistics of the photon radiance. This allow to mutually weight different channels of a multispectral images. Except for a significance level no other thresholds are required.

@INPROCEEDINGS{bruegelmann*92:noise, author = {Br\"ugelmann, R. and F\"orstner, W.}, title = {{Noise Estimation for Color Edge Extraction}}, booktitle = {Robust Computer Vision}, year = {1992}, editor = {W. F\"orstner and S. Ruwiedel}, pages = {90-107}, publisher = {Wichmann, Karlsruhe} }

@inproceedings{förstner91:statistische, author = "Förstner, Wolfgang", title = "Statistische Verfahren für die automatische Bildanalyse und ihre Bewertung bei der Objekterkennung und -vermessung", year = "1991", city = "München", proceeding = "DGK",}

@inproceedings{förstner89:image, author = "Förstner, Wolfgang", title = "Image Analysis Techniques for Digital Photogrammetry", year = "89", pages = "205-221", city = "Bonn", proceeding = "Photogrammetrische Woche 1989",}

The paper discusses the accuracy potential of mono and stereo image sequences. Specifically treated are the effect of image blur onto the precision of image features, the precision obtainable for relative position, orientation and speed of the sensor platform with respect to a given coordinate frame or to other moving objects. The results can be used for designing a multisensor navigation system.

The paper describes a concept for object location, when not only image features but also the model description is uncertain. It contains a method for probabilistic clustering, robust estimation and a measure for evaluating both, inaccurate and missing image features. The location of topographic control points in digitized aerial images demonstrates the feasibility of the procedure and the usefullness of the evaluation criteria.

@inproceedings{förstner*89:object, author = "Förstner, Wolfgang and Sester, Monika", title = "Object Location Based on Uncertain Models", year = "89", pages = "457-464", city = "Bonn", proceeding = "Mustererkennung 1989, 11. DAGM-Symposium",}

A testfield containing a large number of black targets on a white background has been recorded by a digital camera from many points of view. In each digital image, the targets have been located using elementary image processing techniques. Precise coordinates were obtained by matching the targets with artificial masks. The precision of these coordinates was calculated in a bundle block adjustment with self-calibration parameters. The achieved precision amounted to 0.03 pixel, corresponding to 0.8 um in the image plane.

@inproceedings{förstner*88:precision, author = "Förstner, Wolfgang and Vosselmann, George", title = "The Precision of a Digital Camera", year = "88", pages = "148-157", city = "Bonn", proceeding = "ISPRS 16th Congress",}

@inproceedings{förstner87:reliability, author = "Förstner, Wolfgang", title = "Reliability Analysis of Parameter Estimation in Linear Models with Applications to Mensuration Problems in Computer Vision", year = "87", pages = "273-310", city = "Bonn", proceeding = "CVGIP - Computer Vision, Graphics, and Image Processing",}

Feature extraction is a basic step for image matching and image analysis. The paper describes a fast operator for the detection and precise location of distinct points, corners and centres of circular image features. Distinct points are needed for feature based image matching or for trackong in image sequences. A special class of these distinct points are corners, which, beside edges, are the basic element for the analysis of polyhedra. Finally centres of circular features cover small targeted points and holes, disks or rings, which play an important rôle in one-dimensional image analysis. The extraction consists of two steps: window selection and feature location. The speed of the non-iterative operator results from parallelism on the arithmetic as well on the process level. Specifically the operator can be split into arithmetic operations on and between imgaes, convolutions, partly with boxfilters, and finally vector and matrix operations. The operator provides a measure for the precision of the location.

@inproceedings{förstner*87:fast, author = "Förstner, Wolfgang and Gülch, Eberhard", title = "A Fast Operator for Detection and Precise Location of Distict Point, Corners and Centres of Circular Features", year = "87", pages = "281-305", city = "Bonn", proceeding = "Proceedings of the ISPRS Conference on Fast Processing of Photogrammetric Data",}

A new feature based correspondence algorithm for image matching is presented. The interest operator is optimal for selecting points which promise high matching accuracy, for selecting corners with arbitrary number and orientation of edges or centres of discs, circles or rings. The similarily measure can take the seldomness of achieved by maximum likelihood type (robust) estimation for the parameters of an object model. Approximate values have to be better than 1/3 of the size of the image in shift, 20° in rotation and 30% in scale.

@inproceedings{förstner86:feature, author = "Förstner, Wolfgang", title = "A feature based correspondence algorithm for image matching", year = "86", city = "Bonn", proceeding = "ISP Comm. III",}

@inproceedings{förstner19:text, author = "Förstner, Wolfgang", title = "Text zu "A feature based correspondence algorithm for image matching"", year = "19", city = "Bonn", proceeding = "ISP Comm. III",}

@inproceedings{förstner19:feature, author = "Förstner, Wolfgang", title = "A feature based correspondence algorithm for image matching", year = "19", city = "Bonn", proceeding = "ISP Comm. III",}

@inproceedings{förstner19:abbildungen, author = "Förstner, Wolfgang", title = "Abbildungen zu "A feature based correspondence algorithm for image matching"", year = "19", city = "Bonn", proceeding = "ISP Comm. III",}

@inproceedings{förstner85:determination, author = "Förstner, Wolfgang", title = "Determination of the Additive Noise Variance in Observed Autoregressive Processes using Variance Component Estimation Technique", year = "85", pages = "263-274", city = "Bonn", proceeding = "Statistics and Decision, Supplement Issue No. 2",}

@inproceedings{förstner85:prinzip, author = "Förstner, Wolfgang", title = "Prinzip und Leistungsfähigkeit der Korrelation und der Zuordnung digitaler Bilder", year = "1985", city = "Bonn", proceeding = "Photogrammetrische Woche 1985",}

The paper discusses aspects of evaluating the results of digital correlation used in photogrammetic high precision application. The most common correlation techniques are compared with respect to their optimization criteria. Results from practical and theoretical investigations concerning the sensitivity of the methods with respect to deviations of the mathematical model from reality are given. The aim of the paper is to provide some insight into the dependency of the main parameters of digital image correlation on the image texture, e.g. the pixel and the patch size, the quality of approximate values, the influence of unmodeled geometric distortions or of correlated noise. The results are useful for increasing the adaptility of the methods. Key words: Digital correlation, object location, point transfer, quality assessment, least squares, variance, convergency, pull-in range, geometric distortions, correlated noise, matched filter, Wiener Filter, phase correlation, robust estimators.

@inproceedings{förstner84:quality, author = "Förstner, Wolfgang", title = "Quality Assessment of Object Location and Point Transfer Using Digital Image Correlation Techniques", year = "84", city = "Bonn", proceeding = "International Archives of Photogrammetry",}

Rectification of single and overlapping multiple scanner frames is carried out using a newly developed comprehensive parametrix model. Tests with both simulated and real image data have proven, that this model in general is superior to the widely used polynomial model; and that the simultaneous rectification of overlapping frames using least squares techniques yields a higher accuracy than single frame rectification due to the inclusion of tie points between the image frames. Used as control, edges or lines, which are much more likely to be found in images, can replace conventional control point and can easily be implemented into the least squares approach. An efficient algorithm for finding corresponding points in image pairs has been developed which can be used for determining tie point between image frames and thus increase the economy of the whole rectification procedure.

@inproceedings{förstner*84:rectification, author = "Förstner, Wolfgang and Mikhail, Edward M. and Paderes, Fidel C.", title = "Rectification of single and multiple frames of satellite scanner imagery using points and edges as control", year = "84", city = "Bonn", proceeding = "Nasa Symposium on Mathematical Recognition and Image Analysis",}

@inproceedings{förstner83:morphological, author = "Förstner, Wolfgang", title = "On the Morphological Quality of Digital Elevation Models", year = "83", pages = "6.1 - 6.18", city = "Bonn", proceeding = "Proceedings of the ISPRS Comm. III/WG 3 International Colloquium on Mathematical Aspects of Digital Elevation Models, Photogrammetric Data Acquisition Terrain Modelling, Accuracy",}

The refinement of the functional model used for photogrammetric point determination has lead to a significant increase of the accuracy, being about 3-8 μm at photoscale. It is discussed how the functional or the stochastical model may be further refined to compensate for varying, systematic effects and for local distortions which are caused by time-dependent changes of the flight an measuring conditions.

@inproceedings{förstner82:systematic, author = "Förstner, Wolfgang", title = "Systematic Errors in Photogrammetric Point Determination", year = "82", pages = "197--209", city = "Bonn", proceeding = "Proceedings Survey Control Networks, International Federationof Surveyors (FIG), Meeting Study Group 5B",}

The geometric precision of digital correlation can be described by the standard deviation of the estimated shift. The paper shows how the precision depends on the signal to noise ratio, the number of pixels involved and the texture of the object and discusses the choice of a low pass filter which minimizes the variance of the estimated location in order to obtain on optimal sampling frequency.

@inproceedings{förstner82:geometric, author = "Förstner, Wolfgang", title = "On the Geometric Precision of Digital Correlation", year = "1982", pages = "176--189", city = "Bonn", proceeding = "Proceedings of the ISPRS Symposium Mathematical Models, Accuray Aspects and Quality Control",}

@misc{foerstner79.verfahren, author = "Wolfgang Förstner", title = "Ein Verfahren zur Schätzung von Varianz- und Kovarianzkomponenten", year = 1979, }

@misc{foerstner79.verfahren, author = "Wolfgang Förstner", title = "Ein Verfahren zur Schätzung von Varianz- und Kovarianzkomponenten", year = 1979, }

@ARTICLE{rosenfeld79:digital, author = {A. Rosenfeld}, title = {Digital Topology}, journal = {American Mathematical Monthly}, year = {1979}, volume = {86}, pages = {621--630}, }