System approach to disparity estimation

Abstract

A system approach to disparity estimation using dynamic programming is presented. The four step system can calculate a dense correspondence map between a stereo pair with parallel or nonparallel camera geometry. Results are presented with CCIR 601 format stereo images. Zntroduction: Binocular stereo is the process of obtaining depth information from a pair of left and right camera images [l]. Surveys on stereo matching techniques can he found in [2, 31. Many approaches are system oriented like the approach described here. The system consists of four parts (Fig. 1): No. 11 nodes of the first stage and the second stage of the graph represent the possible values of Q and 0, respectively. An edge leads from a node of the first stage, say from w, to a node of the second stage, say to 8, if there exists a 1 round characteristic x = (w, 8, 1/2+q), for which q z 0. The weight of this edge is loglql.