Improved Particle Filter For Video Tracking

Video tracking is a hot topic in the field of Computer Vision, which is widely used in the fields, such as security monitoring, intelligent transportation, human exchange, military and so on. As a bottom part of Computer Vision, the quality of video tracking has a direct impact on subsequent studies. Among the many video tracking algorithm, the particle filter has been paid more attention for the advantage of effectively dealing with the nonlinear and non-Gaussian system. However, many problems, which limited the widespread use of the particle filter, in the application of video tracking based on particle filter has to be tackled. In this paper, we improved the timeliness and accuracy of video tracking to make it satisfy the practical applications by introducing some existed algorithms. The main research and outcomes of this paper are as follows:Firstly, for the high computeational complexity problems in video tracking based on particle filter, we proposed an improved particle filter algorithm based on image compression for video tracking. The method analysis the compressed video frame by particle filter, which can greatly save the time consumption of particle filter in the histogram anlysis, and thus improve the video tracking rate.Secondly, for the IO time-consuming issue of video tracking technology based on particle filter, we proposed a multithreaded asynchronons IO model to solve it. Using multithreaded asynchronons IO model, we can reduce the system IO operation time, such as image display, calibration target and target preservation, thereby enhancing the video tracking rate. Experiments show the improved particle filter based on the compressed image and asynchronous IO, as long as the number of particles less than 350, which does not matter with the size of target, will be able to do real-time video tracking.Finally, the video tracking technology based on particle filter requires a large number of particles to achieve accurate tracking, we propose an improved particle filter by introducing mean-shift correction for video tracking. First,we get a inintial target position using particle filter, and set the goal position in the vicinity of initial position. Then use the mean-shift algorithm to correct the initial position, make the target lies in the target position. particle filter method just needs to make a general judgment to get the position,not requires a lot of particle number, so the improved algorithm not only improves the video tracking rate, but also improves the accuracy of video tracking. Experiments show that the improved particle filter method, after image compression, asynchronous IO and meanshift correction, only need 70 particles to achieve accurate video tracking, and it can do real-time video tracking.The improved particle filter that introduced in this paper can achieve real-time and accurate video tracking, which can bring a high value in social development by put it into social production.