Summary

(in german: Zusammenfassung )

During the last years, three-dimensional building data have increasingly gained importance for practical applications as in architecture and planning. Aerial images are suitable for the photogrammetric reconstruction of buildings. This thesis makes a contribution to automatic building extraction from multiple overlapping images. The reconstruction comprises the extraction of the geometric as well as the semantic object description, i.e. the interpretation of the objects.

The modeling of knowledge about the scene by means of object-, image- and sensor-modeling plays an important role in the extraction from image data. The complexity of the objects as well as their appearance in the images requires the use of a hierarchical building model that contains both a part-of and a specialization hierarchy.

The part-of hierarchary comprises different levels of domain-specific aggregation and enables the access to the semantics of the objects by using the class-specific geometric and topologic properties of the objects and the connectivity relations between objects of different levels.

The specific features of the procedure are the following:

• The use of a hierarchical building model with semantically motivated levels of the part-of and the specialization hierarchy for reconstruction of building componentes and for their aggregation in object space.
• The coherent modeling both in object and in image space. Starting from the image data the image model allows the access to 3d-components of the objects during the reconstruction. Starting from object components the object model allows the access to complex objects during the 3d-aggregation by using semantically related connectivity structures.
• The close interaction of geometric and semantic reconstruction that combines the mainly data-driven construction of model instances and the model-driven verification exploiting the maximum available knowledge.
• Uncertainties are taken into account by statistically modeling the deviation between data and model.

The procedure allows the reconstruction and interpretation of complex buildings of variable geometry, topology and structure.

The algorithm is demonstrated on different image material. It shows promising results, especially in suburban areas.

Details

• Overview
• Building Model
• 3D-Corner Reconstruction
• 3D-Corner Aggregation
• Results

Related Projects

• Semiautomatic 3D-Building Reconstruction
• Semantic Modeling for Building Extraction from Digital Images

For any questions ...

please contact:	Felicitas.Lang
or mail to:	lang@ipi.uni-hannover.de