Research On Local Cost Aggregation Matching Algorithm Of Binocular Vision

Binocular stereo vision technology is an important branch of computer vision,stereo matching is the most critical and most difficult step.Although the scholars have proposed many local matching algorithms,due to the presence of occlusion region,the edge of the disparity map is blurred and foreground fattening.In order to solve the foreground fattening of the disparity map,the key are to find the depth discontinuous region?select the appropriate cost aggregation window and assign appropriate weights.Therefore,this paper proposed an cost aggregation algorithm based on adaptive window and multiple weights.The main works are as follows:(1)Extracted the depth discontinuous region.In order to solve the foreground fattening of the disparity map,we first need to extract the depth discontinuous region.Due to the texture of the original image is more complex,and most of the edges are not real disparity edges.Therefore,this paper first obtained the non-cost aggregate disparity map,and an iterative median filter is used to the disparity map.Then,we extracted the edge of the filter image to obtain the depth discontinuous region.(2)Calculated the cost aggregation window size of the depth discontinuous region.After extracting the depth discontinuous region,we used mean-shift algorithm to segment the original image.The segmentation results are calculated exponentially to determine the window size of the pixels in the depth discontinuous region.(3)Cost aggregation algorithm based on multiple weight.While considering color and distance differences,the results of image segmentation,non-cost aggregated disparity map and left-right consistency check are all introduced into the weight calculation process.To improve the accuracy of weight selection.(4)Limited the range of disparity search and speed up the algorithm.The improvement of window and weight also increased the complexity of the algorithm while improving the accuracy.So this paper proposed to reduce the execution time by narrowing the disparity range.The experimental results proved that the method can not only shorten the running time,but also maintain the accuracy.Finally,the experimental results are given.The average mismatching rate of the initial disparity map is 5.14%,and the average mismatching rate of the optimized disparity map is 4.52%.Compared with other algorithms,the average error of our method is the lowest and the matching accuracy is high in all three regions.The effectiveness of the proposed aggregation algorithm is proved.