Improved binocular stereo matching algorithm based on AD-Census

To improve the accuracy of the traditional AD-Census algorithm, first, bilateral filtering is used to preprocess the image to retain the edge information, and then the algorithm is improved from two aspects: one is to construct a gradient-information entropy bimodal perception model to realize the adaptive adjustment of the window and improve the matching effect in heterogeneous texture regions; the other is in the cost calculation stage, an information entropy-driven adaptive fusion strategy is adopted to optimize the weighted fusion of the AD cost and the improved Census cost. Experiments on the Middlebury platform show that the improved algorithm reduces the average mismatch rate in the global region by 7.36%, 7.65%, and 27.30% respectively compared to the original AD-Census, SGBM, and SGM algorithms, and still maintains the real-time property of the algorithm, verifying its effectiveness.