Building extraction from stereoscopic aerial images

Three-dimensional models of urban objects are widely used in geographic information systems, telecommunications, or defense applications. The classic technique for obtaining such models is stereoscopy. Images are densely matched, and images of above-ground structures are delineated. We propose two semiautomatic methods based on the Hough transform and statistically active models to delineate buildings. The first one delineates rectangular shapes; the second one deals with more-complex buildings. Each one is based on a criterion optimization that takes both photometric and altimetric information into account. Results based on real data show that the first method is robust and that the second one, which deals with a broad range of buildings, seems to be a good compromise between robustness and applicability. (C) 2004 Optical Society of America.