Three-view matching algorithm for multipolyhedron reconstruction using genetic algorithm

In this paper, a novel stereo matching algorithm based on genetic algorithm (GA) is proposed to find the correspondences among multipolyhedron objects in a three-view system. Optimal camera configuration in the approach is used to reduce the feature searching area on the image plane, and GA is then used to refine the matching results. Due to optimal camera configuration, we can achieve a smaller searching area, less ambiguity, and a faster matching process. The features on the second image near the intersection of the two epipolar lines derived from the first and the third images considered as possible triplets are refined using GA. The fitness function for every corresponding triplet is assigned according to the corresponding distance between the feature location and the intersection of two epipolar lines. The consistent connection relationship is then enforced in the evolution process to resolve the ambiguous correspondence triplets. Experimental results show that the 3-D multiple polyhedra in a complex scene can be successfully reconstructed in a three-view system.