Stereo Matching And Virtual Viewpoint Algorithm Based On Binocular Stereo Vision System

Stereo vision is one of the most valuable research fields in computer vision,which is widely used in many fields such as automatic navigation,film entertainment,military and national defense,etc.Stereo matching and virtual view generation are two key technologies of binocular stereo vision system.The speed and accuracy of these algorithms directly affect the performance of the whole system.Stereo matching is the process of calculating disparity between left and right views.Disparity maps combine with camera calibration parameters can obtain 3D reconstruction data of scene.Virtual view image can be obtained by disparity map and camera calibration parameters through virtual view algorithms.At present,the difficulties of stereo vision system include: the accuracy and speed of stereo matching algorithm need to be improved;there are some cracks,holes and overlaps in the virtual view generated by virtual viewpoint algorithm,and the existing hole filling algorithm is difficult to solve these problems.In order to solve these problems,an adaptive window based guided filtering stereo matching algorithm and a hole filling algorithm based on disparity map are proposed in this paper.This paper completes three-dimensional camera calibration,image distortion correction and epipolar line correction,stereo matching algorithm and virtual viewpoint algorithm in computer vision.Based on OpenCV,OpenGL,MATLAB platforms,a whole set of algorithms and three-dimensional data are implemented.The effectiveness and stability of the algorithm are verified by experiments.The main work accomplished in this paper is as follows:1.Investigating the scheme of stereo vision system,studies the conversion relationship between 2D images and 3D images in the field of computer vision,and determines the algorithm flow of the system,which includes five parts,namely,three-dimensional camera calibration,image correction,stereo matching,three-dimensional reconstruction and virtual viewpoint.2.Investigating the stereo calibration algorithms,build a binocular vision experimental platform,and carry out calibration experiments with dual cameras and calibration boards.On OpenCV platform,image distortion correction and epipolarrectification are carried out for left and right views.And the experiments of real scene had been carried out to verified the validity of camera calibration,stereo matching and three-dimensional reconstruction algorithms.3.A guided filtering method based on adaptive window is proposed,and its effect is tested with standard image on the platform of MATLAB.The experimental results show that the adaptive window-based guided filtering method can achieve higher matching accuracy,better robustness and higher peak signal-to-noise ratio(PSNR)than the common methods.4.Based on the aforementioned technologies and triangulation principles,three views of 3D point cloud data corresponding to the actual scene 2D images are given on OpenGL platform,which verifies the effectiveness of the proposed camera calibration and stereo matching algorithm in practical application.5.A hole filling algorithm based on disparity map is proposed to reduce the pixel holes in virtual view.The effectiveness of this method is verified by comparing the results of this method with those of existing hole filling algorithms.