Research On Stereo Matching Algorithms Of Binocular Stereo Vision

Computer Vision technology is one of the most important means to access3D information of the actual space. Binocular stereo vision system is playing an important role in three-dimensional information acquisition,3-D reconstruction and rehabilitation. It is also widely used in meteorological remote sensing, robot vision navigation and3-D display. As the last two steps in binocular stereo vision, stereo matching and3-D reconstruction is the focus of current research. Many solutions have been brought out to solve these problems. However, because of the objective conditions like image texture details, noise and shadowing, one single stereo matching algorithm cannot get a dense disparity map with high accuracy.In this paper, binocular stereo vision system are described, then a new algorithm combined image segmentation with belief propagation are introduced, which could solve many problems that one single algorithm cannot deal with. Image segmentation algorithm is one important tool in image processing. Belief propagation algorithm is the optimization algorithm based on global energy function minimization. Through information propagation between pixels, it will get the highest degree of confidence, while energy optimization function is the most stable. Firstly, segment reference image using mean shift algorithm. After this, pixels with similar grayscale or colors are assigned to the same segmented region, which can meet the assumption that disparity values vary smoothly in those regions. Because of over-segmentation, region boundaries are well preserved without being weakened. Then, get the initial sparse disparity map using one single matching method. Correspond the initial disparity map to the segmented areas, and then each area will share one disparity plane. Apply plane fitting method to get the disparity plane parameters. Repeat this to make it more accurate. Stereo matching is then taken as a problem of labeling energy optimization, which will bring in Belief propagation algorithm. Belief propagation, known as BP, uses message passing mechanism to optimize the labeling energy function, and this will do help in stereo matching. Messages are transferred between segments instead of single pixel point. This makes time complexity related to segments not points and helps enhance time efficiency.After a detailed description of the algorithm, the author of this paper does experiments using existing stereo image pairs and gets a dense disparity map with high accuracy. The effectiveness of the algorithm has been confirmed.