Research On The Application Of Guided Filtering In Binocular Stereo Matching

With the acceleration of artificial intelligence,the application of binocular vision is becoming wider and wider.The ultimate goal of binocular vision is to obtain the three-dimensional coordinates of the object in the scene.In the research of binocular vision,more and more emphasis is put on matching.The mismatched pixels produced in stereo matching directly affect the accuracy of the whole system.Therefore,the research on matching is necessary.At present,there are two main problems in stereo matching:(1)in the weak texture area,the matching between pixels will not be accurate because of the indistinct features of pixels,which will reduce the accuracy of disparity.(2)In the edge region of the image,the effect of stereo matching will be poor,and the phenomenon of edge blur will appear when we get the disparity map.In this paper,two algorithms are proposed for two problems.In this paper,a stereo matching algorithm based on two color space model is proposed.The pixel features are described in CIELAB and RGB color spaces.The traditional one-dimensional pixel features are changed into six dimensions,which makes the pixel features more obvious in the weak texture area,enhances the discrimination between pixels and reduces the mismatch between pixels.The experimental results show that the error matching rate of the proposed algorithm is4.75%,and the effect is improved obviously Obviously.Then,a stereo matching algorithm based on twice weighted guided filtering is proposed.In the cost aggregation stage,the regularization parameters of the guided filter are improved,and different values are provided in different filtering windows,so that the processing effect of the algorithm in the image edge area is improved,and the edge of the disparity map is clearer;at the same time,the flow of the cost aggregation stage is modified,and the cost reorganization is completed by fusing the census transform.It can further enhance the edge preservation of disparity map.The experimental results show that the error matching rate is only 4.61% on Middlebury website.Figure [52] table [12] reference [94]...