Application of Improved SAD-Census Conversion Algorithm in Binocular Ranging

Aiming at the problems that the traditional Census conversion algorithm in binocular ranging system depends too much dependence on the central pixel and takes a long time to calculate the cost, an improved Census conversion algorithm combined with sum of absolute differences (SAD) was proposed. The 8 fixed pixels around the central pixel were selected, the pixel values in sequence were compared to obtain bit string of a byte, and the initial cost of Census and SAD were unified as the initial cost of the corresponding support window by the normalization process. Then corresponding disparity map was obtained to measure the distance. The development experiment was carried out in the Visual Studio2017 integrated environment with C++ language to verify the effectiveness of the improved SAD-Census conversion algorithm. The results show that with the increase of support window, the cost calculation time of this algorithm for high pixel and low pixel graphics is stable at 6.0 s and 0.2 s respectively, which is 1/5, 1/4 and 1/3 of the cost of AD-Census algorithm, traditional Census conversion and SAD algorithm, and its stereo matching speed is greatly improved. When the measured object is 800-2 950 mm away from the optical center of the camera, the absolute value of the relative error between the distance obtained by the algorithm in this paper and the actual distance is less than 5%, and the accuracy meets the ranging requirements.