Research On Binocular-vision-based Tracking And Positioning Of Moving Target In Multi-media

The application of binocular vision techniques on three-dimensional measurement and the moving object tracking and locating is a hot research topic in computer vision, which has been used widely in the military vision-guided, robot visual navigation, traffic management and industrial inspection and other fields. Due to the impact of environment and equipment, how to improve the measuring accuracy, track and identify dynamic objects correctly in multi-media is a difficult problem for practical applications at present.Based on the principle of binocular vision, a system for tracking and locating moving objects in multi-media space is designed in the paper. Firstly, the calibration board and corresponding detection algorithm is designed based on Zhang Zhengyou binocular camera calibration principle, which improves detection accuracy up to sub-pixel level, thus the accuracy of calibration is improved; On moving objects detecting and tracking, since Mean-Shift algorithm doesn't contain object's motion information in the process of object tracking, it easily fails to track the objects, an improved algorithm combined with Kalman filtering is proposed in the paper, which improves the accuracy of detecting and tracking moving objects effectively; Then, Combined the refraction theory of light in the critical surface with the spatial geometry theory, the principle of visual measurement in multi-media is researched, and a mathematical model is constructed and tested in the paper.Finally, the whole system is set up and tested, experiment result shows that this system is precise and stable, and it can realize tracking and extracting 3D coordinates of moving objects in multi-media, meet practical requirements, and it is worthy for using and promoting.