Based On The Particle Filter Target Tracking Algorithm Research And Implementation

Object tracking is always one of the hot research fields of computer vision, which includes machine vision, image processing, pattern recognition, probability disciplines and so on, and it has wide applications in the fields of prospects in intelligent transportation system, monitor video and intelligent robot.This article based on the ChongQing key torch-plan projects: the key technology research and development of the video abstract formation and fast browsing from the embedded WEB video. The key point of the object tracking is the researching of tracking prediction algorithm, such as kalman filter and particle filter algorithm, which are the most popular methods. Moving object which is monitored in video moves randomly, more over the kalman filtering that can only solve problems which are Gaussian and linear is not an ideal choice, so we study the particle filter algorithm which can solve the problem of nonlinear and non-Gaussian. The mainly works and innovations in this paper are as follows:1. The model of moving object and occlusion are discussed in the framework of Bayesian filter in details, and the advantages and disadvantages of Kalman filtering and particle filter algorithm is analyzed and compared. Introducing the particle filter algorithm for single object tracking, our main work is the analysis of particle sampling and re-sampling.2. Particle filter algorithm is improved by combining foreground detection. This algorithm is improved in three aspects by foreground detection: object color's histogram which is the characteristics of object is extracted from the interesting region; accuracy of model is improving by the region of interesting. Secondly the efficiency of sample is improved by restricting particles sampling area in the target zone, which can guarantee the credibility of the particle's diversity, and slow down particle degradation speed. Finally, advanced particle filter can well fit to non-rigid object change of appearance, making the particle adapt to the problem of object tracking under complex conditions better, which greatly improves tracking efficiency of particle filter.3. The method which combines the advanced particle filter with technology about object tracking, achieves a good tracking performance in the field of multiply target tracking. The independent tracking module is developed based on the opencv1.0 vs2005; it lays a foundation for the project development in future.