Research On Mul-target Detectiong And Tracking

Multi-targets detection and tracking technology is of great importance in computer vision. The problem of how to taking target fast and exactly has its researching significance and value in the fields of military, scientific and industry. This subject discussed about technology of image registration, motorial targets detection and multi-targets tracking based on image frames of dynamic background in monocular vision.Firstly researching on global motion estimation of Taylor-based sub-pixel iterative algorithm is carried out. This algorithm has better background motion compensation result for translation, rotation and zoom changes by using affine transformation model with six parameters. Then a new moving targets detection algorithm based on grads is presented after comparing advantages and disadvantages of commonly ones. This new algorithm combines moving information between frames and grads information in one frame to improve the connectivity and integrity of the result of difference detection method. Experiments prove the validity and feasibility to detect moving targets in moving scene of algorithm about background motion compensation. Finally, we analysis the multi-target tracking state and its principle, and study the JPDA algorithm and Kalman filter. By using trace initiation, trace estimation, tracking gate selected, data association,trace combination and trace termination 6 steps to finish target tracking process.Emulation result shows using this method to tracking multi-targets can reach our expected result. However more research on the JPDA algorithm has to been taken in the future.