基于改进稳态匹配概率的立体匹配算法研究

针对稳态匹配概率(Steady-State Matching Probability,SSMP)立体匹配算法在处理视差范围大的测试图中产生的空洞现象以及使用该算法后由于右视差图中的错误视差导致的左视差图中正确视差丢失问题,提出一种基于稳态匹配概率和半全局匹配(Semi-Global Matching,SGM)相结合的立体匹配算法。首先使用SSMP算法求取初始视差图。接着,使用基于爬山法颜色分割的填充准则进行填充。然后使用SGM算法重新获取视差图,将两幅视差图中一致的视差信息填充到经过左右一致性检测后的含有空洞的视差图中。最后,使用SSMP算法中的空洞填充和中值滤波得到精化后的视差图。实验结果表明,改进后的SSMP算法在Middlebury测试平台上第2版本的四组图像的平均匹配误差从5.38%减少到5.23%,第3版本部分测试图像的平均匹配误差从24.7%减少到21.5%,该算法能很好地处理上述问题,有效提高匹配精确度,且具有鲁棒性。

Stereo matching algorithm based on improved steady-state matching probability

Steady-State Matching Probability (SSMP) stereo matching algorithm will generate hole phenomenon in the test charts with large parallax. And the correct disparity will be lost in the left disparity map due to the wrong parallax in the right disparity map after using the algorithm. In order to solve these problems, a stereo matching algorithm based on the combination of SSMP and semi-global matching (SGM) is proposed. First, the initial disparity map is obtained by using the SSMP algorithm. Next, the disparity map is filled with the filling criteria based on the Hill-climbing color segmentation. Then, the disparity map is retrieved by the SGM algorithm. The consistent disparity information in the two disparity maps is filled in the disparity map with holes after the left-right consistency detection. Finally, the refined disparity map is obtained by the empty filling and median filtering in the SSMP algorithm. The experimental results demonstrate that the average false matching ratio of the improved SSMP algorithm decreases from 5.38% to 5.23% on the second version and decreases from 24.7% to 21.5% on the third version of Middlebury testing benchmark. The proposed algorithm is able to deal with the above problems well, improve the matching accuracy effectively and have good robustness.