Research On Stereo Matching Algorithm Of Binocular Vision

With the rapid development of science and technology,artificial intelligence is gradually entering our life,and machine vision undoubtedly plays an important role,among which binocular vision is favored by researchers because it is similar to the two"eyes" of human beings.The accuracy of stereo matching plays an important role in the accuracy of subsequent research and application of binocular vision.Most of the existing methods used a single similarity measure function to calculate the cost of pixels,which makes the pixel features less in areas with less image texture,thus increasing the difficulty of matching and causing low accuracy of disparity map,or the edge contour in disparity map is unclear due to improper processing in edge areas,thus reducing the accuracy of disparity map.In order to solve the problem of low parallax accuracy in weak texture region and edge region,this paper proposes two stereo matching algorithms,which are as follows:The stereo matching algorithm with multi-information cost calculation and significant gradient regularization fused three features:color feature,gradient information and gradient angle,keeped the color and structure of the image and increased the gradient angle to make the matching cost calculation result more accurate;The regularization parameters of classical guided filtering are improved,and the regularization parameter model of guided filtering is constructed by using the edge saliency of saliency map to improve the accuracy of edge region.Aiming at the problem that most algorithms are difficult to describe pixel features in areas with less image texture,a dual color space model is proposed,which makes the feature vector dimension of pixels become six dimensions,thus greatly reduced the difficulty of matching and improved its accuracy;At the same time,the regularization parameters of guiding filtering are adjusted by using the edge retention items constructed by different pixel textures in different windows of HSV color space;Combined with Census transform,the cost is aggregated twice.In this paper,aiming at the problems of stereo matching,the two algorithms have achieved better results through experiments,and the mismatching rates tested on Middlebury website were respectively 4.78%and 4.61%.Figure[42]table[5]reference[84]...