A Noise-Resistant Stereo Matching Algorithm Integrating Regional Information
Feng Huahui, Geng Zhang, Xin Zhang, Bingliang Hu · 2018
Focusing on the problem existing in stereo matching that low-SNR image, such as images collected at night, we propose a novel matching framework based on semi-global matching algorithm and AD-Census. This algorithm extends the original algorithms in two ways. First, image segmentation information as an additional constraint is added that solve the problem of incomplete path and improve the accuracy of cost calculation. Second, the matching cost volume is calculated with AD-SoftCensus measure that minimizes the impact of noise on the quality of matching by changing the pattern of census descriptor from binary to trinary. Results of Middlebury standard test data show that the algorithm significantly improves the precision of matching. In addition, a low-light binocular platform is built to test our method in night environment. Results show the disparity maps are more accurate compared to previous methods.