Research On Stereo Matching Techniques Based On Binocular Vision

Stereo simulates human binocular vision system through binocular camera,acquires the image perception objective environment from different angles,and has broad application prospects in industry,military,medical,entertainment and other fields.Binocular stereo vision involves camera calibration,image correction,stereo matching and 3D reconstruction.In determining the projection matrix of camera,the image correction makes the stereo images of the polar at the same level,the corrected image to find the corresponding image points to calculate the space object point disparity,finally using the sum difference as to build three-dimensional model.Among them,the accuracy of image correction and stereo matching directly affect the effect of 3D reconstruction,and how to improve the matching accuracy while ensuring the efficiency of the algorithm is the focus of which.This thesis will focus on the image correction techniques and stereo matching algorithms in binocular stereo vision.Firstly,the image model of binocular camera is analyzed,and a stereo image rectification algorithm based on feature point matching has been implemented.According to the epipolar line geometric constraint,this method only needs to provide several sets of matching feature points to estimate the foundation matrix and the projection matrix of the stereo image pair without the calibration of the camera.Then,the effectiveness of the rectification algorithm has been verified by using the Syntim team images and lab-capture images.Secondly,a stereo matching algorithm based on edge enhancement and adaptive window is proposed.At first,the Sobel operator is used to extract the edges and merge with the source image.Then,the human visual aggregation rules are used to select the cost aggregation window and give weights for different pixels.Finally,the disparity is refined by median filtering and local disparity statistical histogram.The experimental results show that this method can effectively maintain the edge of the disparity map,but still need further processing on the image detail.Considering the shortcomings of the proposed matching algorithm,a new matching algorithm is proposed.The algorithm establishes the Gaussian mixture model of the image,and calculating by the EM algorithm to obtain the high accurate adaptive aggregation window.On each aggregation window,the minimum spanning tree structure is built which the weight relationship between pixels is more concise.Then,a high precision dense disparity map is obtained by correcting the mismatching pixels from the valid neighbors’ disparity.Finally,the effectiveness of algorithm has been verified by using the Middlebury images and lab-capture images.The experimental results show that the proposed algorithm improves the matching accuracy of images,especially in the depth discontinuous region.