Abstract: For the problem that the gray value of the center point of the traditional Census transform window is easily affected by noise and the global matching accuracy is low,a binocular stereo matching algorithm based on improved Census transform is proposed.First,the template value of bilateral filtering is used to replace the gray value of the center point of the Census transform window.In order to enhance the reliability of the initial cost calculation,Sobel operator and the absolute error of gray level and the matching cost of the algorithm are added for cost fusion.Then the dynamic cross region is selected to establish the relationship between adjacent parallaxes.Finally,the winner-take-all strategy is used to select the best disparity,and left and right consistency detection and guided filtering are used to optimize the disparity map.The experimental results show that compared with the traditional Census transform,the average mismatch rate of the algorithm is reduced by about 46% in the noiseless case and 53% in the noisy case.It is concluded that the improved algorithm has better accuracy and anti-noise ability.

Key words: stereo matching, Census transform, bilateral filtering, cost fusion, parallax map