Research On Key Technique Of Stereo Matching Based On Binocular Vision

Due to the rapid development of artificial intelligence technology in recent years,computer vision has been the focus of researchers.The binocular stereo vision system,which simulates the way of observing the world with human eyes,uses binocular camera to obtain environmental images from different angles,giving the machine the ability to perceive the objective environment.Stereo matching is an important part of binocular vision system.In this paper,based on the analysis of binocular stereo vision and calibration principle,the advantages and disadvantages of several classical stereo matching algorithms are compared and analyzed,the post-processing of stereo matching and disparity optimization are studied and analyzed,and a targeted improved stereo matching algorithm is proposed according to the analysis of classical algorithms.Aiming at the post-processing and disparity optimization in stereo matching,this paper focuses on the analysis of the generation principle of occlusion areas and classifies those occlusion areas according to the different generation principles.The principle of detection and post-processing methods for occluded area are analyzed,and the result of disparity map after post-processing is tested to verify the effectiveness and necessity of disparity post-processing and optimization.An improved Census stereo matching algorithm based on adaptive weight is proposed to solve the problem that the traditional Census algorithm is sensitive to noise and has low matching accuracy in weak texture regions.In the cost calculation stage,the reference pixel value is obtained by the weighted calculation of spatial similarity degree,and the threshold value is set to limit the difference between the reference value and the pixel of the central point,so that the proposed algorithm can judge whether the central pixel has mutation in the cost calculation stage and select the reference pixel value adaptively.In the stage of cost aggregation,the multi-scale aggregation strategy is introduced,the guided filter is used as the kernel function of cost aggregation,and the regularization constraints are added to maintain the consistency between scales.The experimental result of the Middlebury datasets shows that the proposed algorithm is significantly more robust than the traditional Census algorithm,and can reach the performance level of the mainstream classical outstanding algorithms in terms of average Percentage of false match.Finally,a binocular vision system is built based on hexapod robot platform to obtain binocular images from real scenario and preprocess them before matching.The proposed algorithm is used for stereo matching test.Experimental results verify the effectiveness of the proposed algorithm in restoring depth information of those images.