Ein modellbasiertes Konzept zur städteplanerischen Kartierung durch digitale Bildanalyse

There is an increasing need for 3D building extraction from aerial images for various applications such astown planning, environmental- and property-related studies. Aerial images usually reveal on one hand a certain amount of information not relevant for the given task of building extraction like vegetation, cars etc. On the other hand there is a loss of relevant information due to occlusions, low contrasts or disadvantageous perspectives. Therefore a promising concept for automated building reconstruction must incorporate a suffciantly complete model of the objects of interest. We propose a model-based approach to 3D building extraction from aerial images which reveals a tight coupling between a generic 3D object model and an explicit 2D image model. The generic object model employes domain specific volumetric primitives (i. e. building part models) and combination schemes. To cover the gap between 3D object models and 2D image data the image model is employed to predict the projective building appearences in aerial images. We present a strategy for a model-based building extraction based on the recognition-by-components principle and show first experimental results derived from international test sets