Research On 3D Reconstruction Technology Based On Binocular Stereo Vision

With the rapid development of computer vision,people's demand for restoring three-dimensional information has increased.Because it does not require complex light source equipment and has many advantages such as convenience,reliability,accuracy,low cost and popularity,the three-dimensional reconstruction based on binocular stereo vision makes it highly practical value in industrial measurement,virtual reality and robot navigation.Binocular stereo vision is more widely used in real life.How to make the binocular vision system take into account the accuracy and speed of operation,while controlling the cost within an acceptable range is currently the focus of binocular stereo vision research.The thesis focuses on camera calibration,system accuracy and stereo matching in the binocular stereo vision technology.The main work of this thesis is as follows:Camera calibration is the premises of the binocular stereo vision system.The accuracy of the calibration is directly related to the accuracy of three-dimensional information recovery.By analyzing the advantages and disadvantages of the camera calibration technology,the thesis selected Zhang's calibration method with flexible operation and high calibration accuracy as the calibration method,which laid the foundation for the work of this thesis.Because of the problem of the influence of binocular system structure parameters on the accuracy of the system,the structural parameter model of the binocular vision system was established to theoretically analyze the influence of the system structure parameters on the measurement accuracy in the thesis.And the measurement error function of the system structure parameters was deduced.Meanwhile,the simulation analysis and experimental verification was done.The conclusion had a good guiding significance for building a high-precision binocular vision system.It studied the principle of SIFT algorithm.Aiming at the disadvantages of a large number of wrong match in the SIFT matching,the thesis used the cross-filtering idea and the RANSAC idea based on the fundamental matrix to eliminate the wrong match,which improved the accuracy of stereo matching.In order to balance the accuracy of stereo matching and running speed of regional stereo matching algorithm,the thesis presented a matching method based on feature matching and seed region growth based on SIFT combined the advantages and disadvantages of feature matching and region matching.Using high-quality SIFT feature matching pairs as seed points and adopting regional growth method,parallax propagation was performed under the parallax gradient constraint,the uniqueness constraint and the grayscale similarity constraint to obtain relatively dense disparity maps.And experimental comparison was conducted to prove that the speed and accuracy of stereo matching of the algorithm had been improved and had certain practicality.Finally,in order to verify the effectiveness of the presented algorithm,a binocular vision system was designed to realize the three-dimensional reconstruction of the real scene.