Building extraction from aerial imagery using a generic scene model and invariant geometric moments

The automatic extraction of buildings from aerial imagery in an urban environment is the main focus of this paper. Aerial col or infrared images and a digital surface model arc used as the source of information. The knowledge about the scene and the geometry of the objects is represented by means of a generic scene model. The strategy of our approach is to reduce the complexity of the image content by means of different abstraction levels. The extraction starts with a description of the coarse content of the given scene. On the scene level the detection of possible building regions is performed. In the next step the knowledge about the surrounding of a building is used in order to support the detection of individual buildings. Finally these buildings are reconstructed using invariant geometric moments leading to orthogonal geometric models. Results of our approach are given in the paper. They demonstrate its feasibility and limitations. The practical application background is to provide a detailed semantic and geometric description of an urban environment, useful for a dynamic 3D simulation of a disaster. Our work is embedded in an interdisciplinary research project funded by the European Commission.