Fast hierarchical stereo reconstruction

Three dimensional reconstruction of alpine terrain based on stereoscopic views from remote line scanners like SPOT requires robust and fast methods that find correspondences between homologue scene points in the stereo images. This matching of a large number of corresponding pairs of points is the most crucial step in the processing chain, both in computational effort and demands on robustness and accuracy. We propose a novel method coming from statistical pattern recognition that utilizes the advantages of hierarchical image representations. It is verified with a SPOT stereo pair. A comparison with the traditional gray- level correlation based matching philosophy shows better accuracy performance, while the decrease factor of necessary computational effort is of several orders of magnitude. A final discussion lists problems and tradeoffs that arise for the matching step, especially in alpine terrain.