Disparity Map Enhancement based Stereo Matching Method Using Optical Flow

Abstract

Estimating disparity from stereo images is a core subject in computer vision. However, the poorly-textured and ambiguous surfaces cannot be matched consistently using the conventional stereo matching method, in other words, the disparity values exist many errors produced by these noises. To solve these problems, based on the characteristics that the optical flow has good robustness to low texture and deep discontinuity, we propose a novel stereo disparity map enhancement approach to improve the accuracy of disparity values in the low texture as well as deep discontinuity regions, meanwhile generate a high-quality disparity map, through performing efficient fusions between the ELAS disparity map and the optical flow image. Experimental results show that the enhanced disparity map obtained by the proposed approach can decrease the bad pixel rate by 2.6% on average compared with the ELAS method, and display accurate and consistent structures robustly.