A progressive constraint search approach for disparity matching in stereo vision

Abstract

This paper presents a novel disparity matching algorithm (called progressive constraint search, PCS) that emphasizes the use of multiple constraints in such a manner that the required search space at a finer constraint level is essentially reduced by the matching results obtained at coarser constraint levels. A dissimilarity measure called the Error function is used to determine whether a template (target) pixel and a candidate pixel are sufficiently alike to form a matching pair. Progressively, from a coarser level to a finer level, PCS employs different constraints at each level to update the Error function. This coarseto-fine strategy effectively enlarges the separation distance J (on the surface of the Error function) between the correct match and false matches, thereby increasing the matching accuracy. Matched pixels found at coarser levels can also be used to guide subsequent searching at finer levels. PCS not only utilizes many constraints to reduce the search space, but also evaluates and verifies a matching pair by thresholding the value of J. The simplicity of implementation and higher accuracy are the two major advantages of PCS over other models.