Binocular stereo vision is an important research field in computer vision, direct simulation human eyes with the way of scenery, to realize the three-dimensional information of the perception, using two cameras on the same scene from different positions, and then recovering from the depth information. Stereo binocular vision has been widely applied in intelligent robots, automation, medical and many other fields, The study of binocular stereo vision not only has important theoretical significance, but also has important practical value.Thesis for the binocular stereo vision camera calibration and rectify, feature extraction, stereo matching and semi-dense disparity map and other key technology research, to achieve the two-dimensional images of three-dimensional reconstruction. Use the camera calibration board algorithm, and the experimental data of calibration are analyzed, to make the data more robust. Summarizes the current several methods for corner detection, compares advantages and disadvantages with them, so we select the good adaptability Harris corner feature extraction. By using zero normalized cross correlation similarity measure feature points, and preliminary matching, using the uniqueness of matching constraint and disparity gradient constraint, remove the wrong match points to precisely match. Improved Jan Cech and Radim Sara's seed growth method to generate semi-dense disparity map, the experimental data shows that the improved algorithm in matching the precision and speed are improved significantly.To realize three-dimensional reconstruction for the real scene is verified by the feasibility and effectiveness of the proposed algorithm, we constructed corresponding hardware platform and software test experimental system, development environment is the VC++6.0. On real scene images of three-dimensional reconstruction of the data results show that this experimental design and the improved algorithm and reasonable. |