Post-aggregation stereo matching method using Dempster-Shafer theory

Stereo matching is a basic yet important issue in the research of computer vision. A key problem of stereo matching is how to efficiently use the information provided by the neighborhood. In some existing disparity refinement methods, it is observed that the disparity is fused only after having the disparity map, which unfortunately causes the lost of cost information. To make a better disparity fusion, a post-aggregation method based on the Dempster-Shafer Theory (DST) is proposed in this paper to replace the traditional Winner-Takes-All (WTA) strategy. DST is used in the post-aggregation by keeping and processing the aggregated cost in each disparity. The experiment is done with real road scenes and the results show that our method fits various cost functions, and that final disparity error can be reduced compared to WTA strategy.