Binocular Stereo Vision-based Target Tracking Algorithm

Human vision is one of the most important means to acquire information. Computer vision is a resemblance to the visual perception procedure of mankind. Vision with dual-camera, which is often called binocular stereo vision, is a core research area in computer vision. With the fast development of the computer and electronics technology, the real-time binocular stereo vision becomes realistic and it almost satisfies the demand of real-world applications. Binocular stereo vision technology can be used in the detection and tracking of the moving objects, which have important theoretical significance and wide practical value.Based on Marr's vision theory, corner detection, camera calibration and stereo matching, which are all important problems of binocular stereo vision, are intensively studied and discussed in this thesis. Binocular stereo vision technology is then used to solve the problems in objects tracking system. Compared with single vision system, binocular stereo vision technology can acquire some extensive information. The main results can be summarized as follows:The definition and detection method of corner are introduced. Harris and SUSAN corner extracting algorithms are intensively studied. An improved Harris corner extracting algorithm was proposed. By introducing quadratic surface fitting method, we obtained sub-pixel corner coordinate. The experiment indicated that the proposed method is preferable.Camera imaging model was built. Zhang's and Tsai's method, which belong to traditional camera calibration methods, are investigated. The corner coordinate in the calibration model plane was accurately measured with universal tool maker's microscope. By the two calibration methods we obtained all intrinsic and extrinsic parameters of the camera. The experiment showed that Zhang's method is better. This result was verified by the reprojection error map method.The stereo matching algorithms were investigated. Based on three measure function, SAD, SSD and NCC, dynamic programming was introduced to improve the matching speed and the accuracy. The experiments showed that the matching result satisfies the request of objects detection.The binocular stereo vision technology is applied to moving objects detection. The distance information of the moving object was acquired. This extensive information makes the tracking system be more practical in real-world applications.