Stereo Matching Algorithm Based on Joint Matching Cost and Adaptive Window

Abstract

Stereo matching is one of the key technologies in stereo vision. Due to its many problems, it has not been well resolved. Aiming at the problem that the traditional SAD similarity measure function easily causes amplitude distortion, a local stereo matching algorithm combining similarity measure functions is proposed. Firstly, in the traditional SAD similarity measure function, the Census transform is introduced to construct a linear weighted matching cost algorithm for adaptive weights; an adaptive window based on pilot filter is constructed using image structure and color information for cost aggregation; A detection strategy is used to detect matching anomalies, and sub-pixel enhancement and median filtering are performed on the obtained disparity map to obtain a final high-precision disparity map. The experimental results show that the algorithm is effective, the matching accuracy is high, and it has better robustness to the conditions of light distortion and edge information.