A Control Point Model Database for Automatic Exterior Orientation

The Project Group

Karl Heiko Ellenbeck
Thomas Läbe
Lemonia Ragia
Bernhard Weber

Task

To do the exterior orientation automatically, one needs not only control points but control point models which could be found in the images by a program. We are using 3D-wireframe models as control points. Actually we build a database for North-Rhine-Westfalia (34000qkm).

Solution

Today we are on the way from analytical photogrammetry towards the use of digital photogrammetric workstations. In the world of digital imagery those steps which are common to most of all projects should be automated. In nearly every case one needs the orientation parameters of the images. Examples of actual avaible automatic tasks are the relative orientation and the digital aerotriangulation. The automatic interior orientation is already solved, too. To complete the orientation step the automatic link between the ground coordinate system and the image-system has to be established. Today an operator has to do this task. Control-points must be detected in the images. For best orientation results the size of the control points should fit to the measuring mark of the photogrammetric instrument. To automate the point detection procedure in digital images larger sized structures are needed because the automatic procedures need a certain amount of signal pixels. If we use signalized control points we also need larger sized signals. Here we have the advantage that we could use the same template for many images and control points. But putting the signals on the ground by hand is costly and not adequate in urban regions. Another proposal is the use of old image-patches of natural control points. One has to do a matching between the mask and new images with unknown exterior orientation. Problems with different vegetation and illumination occur here. When the new images have another scale or the perspective has changed, this approach seems to fail because we have only 2D-structures as a reference.

In our work the use of 3D-control point models is described. The models are time invariant to avoid the problem of different vegetation. Roads or buildings are such structures. Regarding buildings one finds out that they mainly consist of straight lines. These lines could be easier modeled than curves of roads. Our approach uses 3D-wireframe models of buildings as ground control features. With the help of approximate orientation values the models are projected in the images to avoid the problems of the perspective view. A pose clustering is used to search the position. Because of the redundancy a false position of control point models could be detected and corrected. After this correction a robust spatial resection fits all the modells in an optimal way using the correspondence of the 3D modell edges to the image edges. After that a selfdiagnosis analysis verifies the result with respect to precision and sensivity. This enables the automatic procedure to decide whether the determined orientation parameters are acceptable or should be rejected.

For the production of the orthophoto map 1:5000 of the state of North-Rhine-Westfalia (about 34000 square km) an old time invariant database of control points which contains of single roof points exists. To build up an automatic batch process for the digital orthophoto production the data base has to be changed to 3d-wireframe models. For each control point model one needs 3D-edges of buildings or groups of buildings which are time invariant.

We are building a new database by a blockwise bundle adjustment of the at last used areal images of 1:12000 scale with 60% forward and 30% sideward overlap. The old control points are used for orientation. The measuring is done on an analytical plotter Planicomp P3 which is additionally equipped with CCD-cameras. The image patches (512 x512 pixel) of the old and proofed houses and of new control models are scanned.

The wireframe models are measured in the digital image patches. In the program PhotoCad an operator has to move a wireframe-modell aproximately at the position of the real building. The length, height and width are chosen, too. Then an automatic fitting is started. A robust estimation optimizes the correspondence between the image edges and the model edges of every image patch. The result is the position and the size of the building in 3D. The result was not measured only in a stereo modell, but was computed out of the 3 to 6 image patches which do the block connection. This data is then stored in a database. So the input of the automatic exterior orientation is a result of image processing technics.

The Survery Department Bonn will use the new database in practice for the orthophoto production process. The images will be scanned with the help of operators. All other tasks including the automatic exterior orientation could be done offline. Only the success of the production process must be checked. So our approach offers the possibility to do the orientation process fully automatically for the main part of North-Rhine-Westfalia.

Literature


tl - 05.03.98