Bayesian inference based high framerate stereo matching and its application in robot manipulation
Yingnan Wu, Cancan Zeng, Jingyu Zhang, Gaobo Xiao, Mingjun Ren · 2019
In stereo matching, the trade-off between the matching accuracy and computation complexity is still an important issue. This paper presents a Bayesian inference based multi-scale weighted voting framework to address the problem by enhancing the accuracy of local stereo matching. The method utilizes the rapidity of the local methods to construct a disparity space with scale information and makes use of the complementarity of the disparity in different scales to find the best disparity distribution by Bayesian inference based weighted voting. The method is generic that can be utilized to arbitrary local stereo matching algorithms and is capable of dramatically improving the matching accuracy without much scarification of the computational efficiency. It is shown in a series of comparison tests that, with the use of the proposed method, the simplest block matching method can be optimized to reach both the accuracy and the efficiency better than several the state-of-the-art real-time stereo matching methods. The proposed method is then integrated into a robot manipulating system to validate its effectiveness.