Stereo Matching Algorithm Based On Mst And Spatiotemporal In Stereo Video

Along with the development of computer science, people no longer meet the two-dimensional visual information, and begin to expand to the3D world, stereo vision has played an important part in modern life correspondingly. Stereo vision has been applied in many fields, including3D reconstruction, object detection, Robert vision, virtual reality, aerospace and so on. Research on stereo vision has become an important branch in computer vision.This thesis focuses on the stereo matching technique, which is the core of stereo vision. Stereo matching is utilized to obtain the disparity value of stereo images, and receive the depth information. The thesis introduces stereo matching algorithm systematically, including the camera calibration and epipolar geometric constraints, the procedure and classification of stereo matching, the accuracy and efficiency of different algorithms. Based on the theory, we proposed new improved method.The new algorithm is an efficient method based on the combination of MST (minimum spanning tree) and Oll (Orthogonal integral image) technique. Firstly, Weber local descriptors (WLD) for each color channel from current stereo pairs are extracted, and the raw matching costs between the images are initialized by WLD. Then raw cost is aggregated by MST based global method, for MST can give a natural image pixel similarity measurement metric and provide a specific support region for each pixel to confirm the integral support region. Multi-thread mode is utilized in this thesis to improve the algorithm efficiency, and reach satisfactory result.A new stereo matching method in stereo video is proposed in this thesis. The new algorithm is based on the MST and spatiotemporal, three-frame subtraction is used to determine the motion region between adjacent frames. The disparity of current frame with the renewed disparity and the one for last frame is confirmed. The proposed approach has been demonstrated with virtual and real stereo image sequences, the accuracy is improved along with the efficiency.The proposed stereo matching algorithm is tested with VS2010and opencv, dynamic link library is generated to make the algorithm modularized. An experimental platform is made to achieve the disparity video, the stereo video is used as the input, and disparity video is extracted to realize the3D display.