A STEREO MATCHING ALGORITHM FOR URBAN DIGITAL ELEVATION MODELS
Caroline Baillard, Olivier Dissard · Photogrammetric Engineering & Remote Sensing · 2000
This paper describes a stereo matching algorithm dedicated to complex urban scenes. It relies on successive complementary matching steps, based on dynamic programming. Intensity edges of both images are matched, which produces piecewise continuous 3-D chains. This provides a description of the scene structure containing the highest elevation of most height discontinuities. Then, the interval pairs defined by the matched edges are linked hierarchically, by a radiometrically constrained process followed by a geometrically constrained one. The novelty of the approach lies in the use of several successive steps appropriate to different kinds of pixels. It provides dense disparity maps with less noise, while preserving discontinuities, a characteristic of urban digital elevation models. The method is reliable and fast, and is robust to image variability. Perspectives within an industrial production context are discussed.