Research On Visual Tracking System Of The Air Moving Target Based On Particle Filter

The technology of visual tracking is a new direction of object tracking in recent years, which compares to the radar tracking it provides more details, so it is easier to detect and distinguish the object. It has lots of applications both in military and civil use, such as tracking and intercepting the rivalry object, the security surveillance for the important place or device, video navigation, etc. The visual tracking is a multiple technology, which integrates technologies of artificial intelligence, computer vision, pattern recognition, graphic procession and so on. Because of the different requirements of real-timing, accuracy, stability and recognizable for the practical application, no one algorithm could fit all situations. So the key point to judge the quality of the algorithm is how to develop the best method according to the practical requirement.The research object based on missile tracking. So the important point of the tracking algorithm is the real-timing and accuracy. Because of the existence of the background noise and device noise, it is hard to get the real situation outputs. The chief work is to resist and eliminate the noise and guarantee the quality of real-timing to improve the reliability.The paper is divided to three parts. The first part researches on the object detection, some common algorithms in visual object detection are introduced, and have some comparison experiment on frame difference algorithm, background subtraction algorithm, and also use OSTU law to get the best threshold which increase the accuracy of the two detection algorithms, and then introduce the improved frame difference algorithm and three frames difference algorithm, and propose the incorporation of the two dimension projection algorithm and three frames difference algorithm to detect object and design the wave-door. Part two discuss the usual filter technology such as standard Kalman filter and extended Kalman filter, and then analyses the particle filter which is based on Bayesian estimation and Monte Carlo method. a new sampling algorithm to improve the particle filter is proposed. On the basis of the research before, a new tracking algorithm whose frame is particle filter which integrate the infrared signature of the object is proposed in the third part, make experiment of the EKF,PF and NPF on MATLAB 7.0 and then analyse the results. The experiment results shows that the algorithm not only achieved the goal of accurate tracking but also enhanced the robustness of the tracking system. The last part makes flying object tracking experiment with the algorithm above on Visual C++ 6.0 and the result shows the feasibility and validity of the improved particle filter algorithm.