Stereo Matching Using Segment Shape as Constraint
Yasuo Seki · Systems and Computers in Japan · 1990
Abstract This paper describes the stereo matching algorithm using the shape information of the edge segment as the constraint. The authors have been studying 3‐D information reconstruction by stereo vision from the viewpoints of the accuracy of correspondence and the versatility of the algorithm. A method was proposed in the course of study where the edge segment is used as the basic unit of the correspondence, and the composite edge segment is used as the real unit of correspondence to cope with the incomplete image processing. Promising results were obtained by the proposed method, but the following problems remained: (1) a heuristic method is employed for the case of more than one candidate, where the candidate with the depth closest to the average depth of the scene (called average‐depth‐oriented heuristics), which results in the restricted range of applications; (2) for simplicity, the correspondence is determined based on the adjustment only among partial contradictory candidates with the result not always being adequate. This paper aims at solution of those problems with the following points. (a) Instead of the forementioned heuristics, a method employing the shape information of the segment as the constraint is proposed which is made possible by the segment method. The direction vector sequence difference is considered as a stable measure for the difference of shapes in the left and the right frames. (b) The hierarchical clique method is proposed as the method of correspondence, where the contradictory relation between candidates is considered globally. (c) The effectiveness of the algorithm is demonstrated by a simulation experiment.