Research On 3D Reconstruction Algorithms Based On Stereo Matching

Three dimensional reconstruction based on binocular stereo vision has attracted extensive attention because of its simplicity and low cost.As the core technology,stereo matching has become the focus of researchers.Aiming at solving the matching problems at low-texture regions and disparity discontinuous regions,the paper has carried out intensive study and proposed the corresponding solutions.Based on the Middlebury standard library and the system established in this thesis,a large number of experiments have been carried out to prove the superiority of the proposed algorithm.The results show that the proposed algorithms can effectively and accurately solve the above problems,while improving the accuracy of stereo matching.The main works and innovation of this paper are introduced below:1.A dynamic programming stereo matching algorithm based on ground control points and prediction points is proposed.Firstly,the ground control points are obtained from three local methods,which can improve the density of ground control points.Then,the concept of prediction points is put forward.The prediction points are separately extracted from low-texture and high-texture regions where different local methods behave differently.At last,ground control points and prediction points are respectively used to guide dynamic programming in strong constraint and weak constraint,which realize the accurate stereo matching.The experimental results based on standard library images and real images show that the proposed algorithm can effectively improve the matching accuracy of all texture regions.2.In order to solve the mismatching problem at disparity discontinuity regions,a novel disparity refinement method based on edge registration is proposed.Firstly,left-right consistency check is used to detect unreliable points and the unreliable points are filled and filtered.Then,edges are respectively extracted from the reference image and the filtered disparity map and then these edges are registered.Thirdly,based on the registration results,correct edges are extracted and then refined.Finally,the corrected edge is diffused to locate the correction region,and the least square method is used to correct the disparity in the correction region.The experimental results show that the corrected disparity map is greatly improved in accuracy.3.A complete 3D reconstruction system based on binocular stereo vision is implemented.The camera calibration,stereo matching,epipolar rectification,point cloud reconstruction and other steps are completed for constructing the system,in addition,the proposed stereo matching algorithm and disparity refinement algorithm are applied to the system,which can improve the precision of the system,finally get the better effect of 3D points cloud.