A Multi-Constraint Strategy Algorithm for Stereo Matching

Abstract

Abstract An improved multi-constraint stereo matching strategy is proposed herein. First, the original reference images are downsampled to reduce the matching scale. Absolute difference and rank transform are used as the matching cost. We establish the associated disparity map and the matching cost matrix to constrain the matching process. We use the associated disparity map to apply a smoothing constraint on disparity values. We then apply a uniqueness constraint using the associated disparity map and the matching cost matrix. We also apply a continuity constraint to improve the efficiency of the algorithm. Second, for the downsampled disparity map, the disparity map of the original image is obtained by interpolation and filling. Lastly, the final disparity map is obtained using a sub-pixel refinement strategy. The experimental results demonstrate that the algorithm is obviously superior to the traditional method. In addition, a multi-constraint matching process may be extracted as a framework for combination with other algorithms. Compared with the original algorithms, the fusion algorithm improved matching accuracy by 3% on average.