Omni-Directional Semi-Global Stereo Matching with Reliable Information Propagation

High efficiency and accuracy of semi-global matching (SGM) make it widely used in many stereo vision applications. However, SGM not only struggles dealing with pixels homogeneous area, but also suffers from streak artifacts. In this paper, we propose a novel omni-directional (OmniSGM) cost volume update scheme to aggregate costs paths along all directions encourage reliable information propagate across entire image. Specifically, perform four tree structures, namely trees the left, right, top bottom root node, then fuse outputs obtain final result. The contributions on each can be recursively computed leaf nodes ensuring our method has linear time computational complexity. Moreover, An iterative is proposed using aggregated last pass enhance robustness initial cost. Thus, useful more likely long distance handle ambiguities low textural area. Finally, present an efficient strategy disparities stable minimum spanning (MST) for disparity refinement. Extensive experiments Middlebury KITTI datasets demonstrate that outperforms typical traditional SGM-based aggregation methods.