PMSGM: PatchMatch Semi-Global Matching for Efficient Stereo Correspondence

Abstract

Semi-Global Matching (SGM) is a robust and widely used method for stereo correspondence because of its good trade-off between disparity image quality and computation efficiency. The computational complexity of the previous works are proportional to the disparity range, thus an efficient SGM-based solution is still very challenging especially for the increasing demand of large disparity stereo matching. This paper proposes a PatchMatch Semi-Global Matching (PMSGM) algorithm which significantly reduces the number of candidate disparities by means of the PathchMatch spatial propagation scheme. The evaluation results on KITTI2015 training dataset demonstrate that, the proposed method can achieve competitive disparity accuracy with approximately 5x and orders of magnitude efficiency improvement compared with the original Semi-Global Matching algorithm and the other PatchMatch-based stereo matching algorithms respectively.