The Study And Implementation Of Parallel Binocular Vision System

ABSTRACT:Binocular vision is one of research hotspot in the field of computer vision, and it has important practical value and broad development prospect, such as video security monitoring, medical diagnosis, intelligent transportation, aerospace and so on. Compared with monocular vision, binocular vision not only has a wider field of view, but also can acquire depth information of the target object through directly imitating the mechanism of human binocular vision. In recently years, with the rapid development of microprocessor and integrated electronic technology, even the binocular vision system which constituted by complex algorithms can basically meet the requirement of real-time, so the potential of binocular system remains to be further explored.Based on the background of video security monitoring in ethnic minority area, the system has been set up and it mainly includes these parts:moving target detection and tracking, target matching under different field of view, and obtaining depth information of the tracked object. According to the overall performance of the system, these results are achieved:1. The system can communicate between PTZ and computer should be accomplished, including communication initialization, automatic scanning and manual fine-tuning of PTZ, PTZ’s position (i.e. horizontal and vertical rotation angle) displayed in real time and so on. In the early stage of the system, camera calibration is an important part for the whole research. Here, calibration board is used to calculate the internal and external parameters of the two cameras.2. Both of the target detection and tracking algorithms which based on static and dynamic background are studied, because cameras would rotate with PTZ, the target detection and tracking algorithms should meet dynamic background situation. Camshift algorithm which developed from Mean shift algorithm could adapt to the changing context to effectively track moving object, especially the algorithm can solve problems of partial occlusion and distortion of object, meanwhile, background difference method is used to realize automatic selection of the moving object. In the paper, Camshift tracking algorithm combined with effective PTZ control strategies make cameras successfully capture and lock moving target.3. In binocular vision system, due to difference in relative position of the two cameras, how to ensure cameras to track the same object and accurately extract feature points will determine the accuracy of depth information, here, SIFT algorithm is adopted by studying many image matching algorithms, SIFT can identify the target under different conditions of illumination and has strong robustness.Finally, we acquire depth information on the basis of parallax principle, the experimental data show that errors within the allowable range, meanwhile, the result prove good function of the whole system. In the recent years, target tracking algorithm based on sparse representation and multi-cameras coordination mechanism are put forward, the two sides provide a feasible manner to further improve the performance of real-time and robustness.