Research on Adaptive Stereoscopic Matching Algorithm Based on SGM

SGM (Semi-Global Matching) algorithm is a binocular stereoscopic matching algorithm that uses the traditional Census transform for cost calculation. The traditional Census transform uses a fixed window size and relies on the center pixel of the window, which is susceptible to noise interference, resulting in reduced matching accuracy in areas with weak textures and discontinuous parallax. Therefore, this paper introduces dynamic adaptive matching windows based on the traditional Census algorithm, and uses different window sizes for different gradient intervals to solve the problem of using a single window in the traditional algorithm, and introduces comparison pixels instead of window center pixels to solve the problem of dependence on window center pixels. Experimental results show that in the test of the Middlebury dataset, the evaluation error rate of the proposed algorithm is $\mathbf{3. 9 3 \%}$ and the global error match rate is $5.34 \%$, which is $4.09 \%$ and $4.37 \%$ lower than that of the traditional algorithm, respectively, and has high accuracy.