Feature extraction of gable-roofed buildings from multi-aspect high-resolution InSAR data

Abstract

The achievable spatial resolution of state-of-the-art synthetic aperture radar (SAR) sensors enables the analysis of urban areas. The appearance of buildings in magnitude images is governed by effects of the inherent oblique scene illumination, such as layover, radar shadow and salient lines of bright scattering caused by direct reflection or multipath signal propagation. For example, in urban residential districts often salient pairs of parallel lines of bright magnitude are observed at locations of gable-roofed buildings. The first line (closer to sensor) is due to direct reflection of planar roof parts orientated toward the sensor. The second line can be related to signal caused by a dihedral corner reflector between ground and building wall. In this paper an approach is presented aiming at reconstruction of gable-roofed buildings by knowledge based analysis considering the mentioned SAR-specific effects. First, line and edge primitives are segmented and grouped to parallel line pair objects. Then for each of these objects geometrical and radiometrical features are extracted in the InSAR images. Based on the interferometric elevation data in the adjacent area of the primitives the projection from slant range into ground range geometry is done. After geocoding, building hypotheses are built from the fused set of extracted primitives from both aspect directions. The estimation of the building height is carried out by two complementing methods: One is based on the extracted geometric parameters and the other on interferometric height data. The reconstruction results are quantitatively assessed by using a high resolution LIDAR surface model as ground truth data.