基于视差与灰度双层支持窗的立体匹配算法

针对传统局部匹配支持窗难以利用空间、灰度距离远的像素信息的问题,提出一种基于视差和灰度的双层支持窗立体匹配算法。该算法根据视差图获得第一层视差支持窗,在视差支持窗内配合参考图RGB颜色灰度值获得灰度相似子支持窗;随后以视差窗、灰度窗、中心像素为优化路径,通过类动态规划算法优化聚合匹配代价;最后采用WTA策略选取最佳视差,更新所有视差,并不断迭代优化视差图直到视差收敛。经过Middlebury 平台的测评,新算法的平均误匹配率为5.15%。

A Stereo Matching Using Double Layer Support Windows Based on RGB Map and Disparity Map

Aiming at the problem that traditional local matching support window is difficult to use pixel information with long distance and gray distance, a stereoscopic matching algorithm using double layer support windows based on disparity map and RGB map was proposed in this study. The algorithm obtained the first layer disparity support window according to the disparity map, and obtained the gray scale similar sub-support window by matching the RGB color gray value of the reference picture in the disparity support window. The aggregation matching cost was optimized by class dynamic programming algorithm with parallax window, gray window and central pixel as optimized paths. Finally, the WTA strategy was used to select the best disparity, update all disparities and iteratively optimize the disparity map until convergence of the disparity map was achieved. After evaluation by the Middlebury platform, the average mismatch rate of the new algorithm was 5.15%.