Building Detection and Footprint Reconstruction from Single Vhr Sar Images

Spaceborne very high resolution (VHR) synthetic aperture radar (SAR) images with meter and sub-meter resolution permit to extract information from urban areas at the level of individual buildings. In order to exploit this information for various application scenarios (e.g. emergency response after natural disasters), robust building detection and reconstruction methods are essential. Different techniques for building detection and reconstruction from VHR SAR have been proposed in the literature [1, 2, 3]. They mainly rely on the availability of multi-modal data which implies multiple acquisitions with different viewing con gurations (changed incidenceand/or aspect angle). However, with the available technology those data cannot be acquired with a single pass, but only after different passes. Thus, the timely availability of such data is not guaranteed. This is of major concern for damage assessment purposes, which exploiting the key advantages of SAR data (independence on the weather conditions and the solar illumination) is a prime application scenario for VHR SAR imagery. In [4], we presented a method for building detection from single detected VHR SAR images, which is suitable for the data acquired from currently operational spaceborne VHR SAR missions. It is based on the extraction of a set of primitives and on their composition to building candidates based on the semantic meaning of the extracted primitives. The semantic meaning is evaluated as the degree of membership of a primitive to a certain scattering class (e.g. facade, at roof), which allows one a quantitative evaluation of the reliability of the derived building candidates. This additional information partially compensate for the low number of building features detectable in single VHR SAR images.