Research On Stereo Rectification And Stereo Matching Algorithm

Binocular stereo vision is an important branch of computer vision, which has been widely applied in various fields such as object recognition, virtual reality and aerospace. The rectification and matching for stereo image pairs is one of the key techniques for stereo vision systems, so whether its efficiency or not decides the further developments and applications of stereo vision.The rectification for stereo image pairs is an efficient solution for decreasing the computional complexity of stereo matching. This paper proposed an improved rectification algorithm in order to solve the missing point problem for objective images produced by conventional recfication techniques. The proposed algorithm makes use of external and interior parameters of stereo camera to rectify stereo image pairs. In this technique, the pixel coordinates of objective image are first decided, and its counterparts in the original image are available from the imrpoved algorithm. Then, the pixel values of original image are assigned to the corresponding ones of the objective image. Therefore, the reversed thinking is applied in the proposed algorithm to eliminate the missing points existing in the objective images. Moreover, the real images are used to evalulate the proposed algorithm, and the experimental results reveal the efficiency and effectiveness.Since the conventional window-based local stereo matching algorithm endures low matching precision, this paper presents an efficient intensity weighted matching algorithm for stereo vision applications. The proposed algorithm makes use of intensity weighted values instead of grayscale value for each pixel in the support window in order to improve the quality of disparity map. The idea is derived from the bionics fundamentals, namely retinal center detector commonly permits high resolution in the interested regions, whereas others do not require the identical resolution. Therefore, this algorithm enables the intensity weight of the kernel pixel to be the highest in the window. With the proposed algorithm, the disparity map is improved significantly with little incremental computation cost compared with the fixed-window (FW) SAD algorithm. Experimental evaluation is performed using benchmark stereo pairs. The simulation results demonstrate that the proposed algorithm can achieve bad pixel percentage of disparity map lower than the SAD algorithm.