BUILDING RECONSTRUCTION FROM INSAR DATA BY DETAIL ANALYSIS OF PHASE PROFILES

The improved ground resolution of state-of-the-art synthetic aperture radar (SAR) sensors suggests utilizing SAR data for the analysis of urban scenes. In high resolution SAR data of built-up areas, especially bright lines are a distinctive building feature. They are often caused by double reflection between ground and building wall.The appearance of such features depends on the illumination geometry, therefore the location of different building walls can be inferred from multi-aspect SAR imagery. Previous studies showed that especially the fusion of orthogonal viewing directions is beneficial with respect to the generation of rectangular building footprints. However, a comparison with cadastral data revealed that the footprints were often oversized. As a consequence, the building height estimates were too low, due to the presence of ground pixels inside the footprint hypotheses. This problem can be solved by investigation of interferometric phase profiles of layover and building roof areas.For this reason an algorithm was implemented to simulate interferometric phase data by taking into account that a mixture of several backscatterers may contribute to the phase value of a single range cell. These simulations are based on intermediate 3D reconstruction results (i.e. building hypotheses). For this purpose, the geo-position of the building hypotheses and the sensor geometry have to be taken into account. By comparison between real and simulated phase profiles contradictions between the reconstruction hypotheses and the measured InSAR data become obvious. These contradictions steer the iterative improvement of building footprint and height estimate. The assessment of the final building reconstruction results uses cadastral building footprints as 2D information and a LIDAR DSM as 3