Research And Implementation Of Stereo Matching Algorithm In Binocular Vision

Binocular stereo vision is an important branch of computer vision, which directly simulates the human visual system to obtain three-dimensional information of the real scene from two-dimensional images, which is widely applied into many fields, such as the robot tracking, video surveillance, industrial inspection, virtual reality,3D video, and so on. A stereo vision system includes four primary sections:camera calibration, image rectification, stereo matching and three-dimensional reconstruction. Stereo matching is the core and difficulty of the system, the accuracy and efficiency of the algorithm directly affects the binocular visual performance in practical applications.So far, solutions emerge in endless for the technical problems of weak texture, occlusions and disparity discontinuities. But how to improve stereo correspondence efficiency without the loss of matching accuracy remains a challenging research direction. This paper focuses on binocular stereo vision system framework, in-depth study of the core theory of stereo matching. In general, the stereo matching algorithm is divided into two categories:local and global stereo matching algorithm stereo matching algorithm. Through these classical algorithms which belong two types of algorithms in-depth research and analysis, and compare the results of experiments to measure the advantages and disadvantages of two types of algorithms, in this paper focusing on the local stereo matching algorithm boot filtering algorithm improvements.An improved algorithm is proposed based on cross-scale cost aggregation for stereo matching because the existing local stereo matching algorithms can not have both accuracy and speed in the current. Firstly the matching cost volume is computed by the intensity and gradient algorithm. Then building each scale images by a Gaussian pyramid, the matching cost volume is aggregated by guide filtering and on the basis of guide filtering aggregates the matching cost volume of different scales by the cross-scale model. Finally, using the patch matching algorithm instead of the traditional winner take all algorithm to compute disparity and using fast weighted median filtering instead of traditional weighted median filtering to refine disparity. The experiments by the Middlebury stereo matching test platform show that the algorithm can rapidly obtain the disparity map and also improve the matching accuracy, our improved algorithm surely has effectiveness and superiority.