Research Of Dense Disparity Based On Global Homography Algorithm

With the constantly development of computer and digital,computer vision technology has being widespread attention,and become a hot research field.Binocular stereo vision is an important branch of computer vision,and its application field has been widened.Binocular stereo matching is the key technology of binocular stereo vision.Its efficiency and accuracy of matching influence the speed and effect of three-dimensional reconstruction directly.It has an important influence for computer vision development for solving this problem.Therefore,it has important significance for the research of binocular stereo matching.This thesis presents a binocular stereo matching algorithm based on global homography method,the result of the algorithm is dense disparity map which can meet the needs of three-dimension reconstruction,as the same time,the algorithm can adapt the selection of disparity range.First of all,correcting the images of binocular stereo matching,and using SURF feature matching to obtain matching points set,then estimating multiple times of matching points set to obtain the mapping matrix which can adapt to the majority matching points,and taking the best estimated mapping matrix as the homographic matrix.After that,for each pixel,computing horizontal difference value between reference image and target image using the homographic matrix,and then taking the maximum value as the homographic constraint value to constrain the search range of disparity space in global dynamic programming.Otherwise,computing the matching cost according to the disparity search range within cost matrix,and calculate the minimal path from the matching cost,the path numerical is the disparity value within cost matrix,then computing all pixels of image and obtain dense disparity.Finally,using the median filter and mean shift method optimize the dense disparity.The matching algorithm can match the image pair with unknown disparity range through the calculation of the homography constraint.At the same time,it can improve the overall efficiency and accuracy of dynamic programming through the search range of disparity space that is certain by homography constraint.The paper experiment uses standard image that provides by Middlebury test platform and a binocular vision camera images.The experimental results show that the homography constraint can determine the disparity scope of the image directly and improve the efficiency of the global matching algorithm and the accuracy of the disparity search range.