Multi-step stereo matching algorithm based on image fragment and confidence propagation

This paper presents a multi-step stereo matching algorithm that can be applied to multiple scenes. To adapt to different application scenarios, the algorithm divides the stereo matching process into three steps: point, fragment, and plane. First, the texture points of an image are extracted and the stereo matching (point disparity) of these points is performed using the improved self-aware matching measure (SAMM) algorithm. Then, according to the edge information of the image, a smooth region is divided into different fragments in the horizontal direction. The disparity estimation of the smooth region (segment disparity) is obtained through the confidence propagation of disparity values of texture points in the fragments. Finally, based on the similarity of plane disparity, a disparity map is generated using the disparity refining algorithm (plane disparity), and a final high-precision disparity value is obtained. The experimental results show that the proposed algorithm has high operational efficiency and accurate disparity estimation. Moreover, the algorithm may be adapted for more application scenarios.