A knowledge-based 3-D building reconstruction from single very high resolution SAR images

Abstract

The modern advanced airborne Synthetic-Aperture-Radar (SAR) sensors are capable of achieving spatial resolution below one meter. Based on the very high resolution (VHR) SAR data, a much finer level of the structural analysis is possible, which provides additional information for the interpretation and reconstruction of mad-made objects in urban areas. In this paper, we focus on analyzing signature of rectangular buildings in VHR SAR images to provide prior knowledge for 3-D reconstruction from single VHR SAR data. A decomposed model is proposed which not only provides a comprehensive set of expressions linking building dimensions to image features but also facilitates the analysis of building appearance under different aspect angles. Based on the recognized effects, a new knowledge-based algorithm is proposed to reconstruct buildings from single-aspect VHR SAR images. The applicability of the approach is verified by the 3-D reconstruction of buildings from real VHR SAR data.