Research On Target Tracking Based On Particle Filter

Video tracking is an important branch of the video interactive technology. It hasa very widely applications in human-computer interaction, target recognition, robotvision and video surveillance. Video tracking is to detect whether there is a target inthe image sequence. If it exists, we extract the target and tracking. The focus of targettracking is how to accurate detection of target, accurate and stable tracking it in avideo.In the target detection, the characteristics of the target object included in thecolor, shape, motion information, contour, etc. It is generally characterized by a singletarget to filter out video background noise interference. And then it extracts the targetto be tracked. But sometimes there are many complications liking as similar colors inthe background of the target, the target interference motion and stationary targetocclusion, it is difficult to filter out all the interferential noise only by relying on thecharacteristics of individual target. It should remove the interference noise by amulti-feature fusion. Multiple integration features improves the accuracy of target thatcan be effectively extracted.In the target tracking, the traditional tracking method can trace to the target, butit is not good tracking stability. Therefore, this paper uses tracking method underparticle filter framework. It can improve the stability of the target track through themulti-particle tracking.In order to solve the problem of low robustness in the tracking based on particlefilter with improved Camshift. This paper uses the color feature and motion trackingfeatures as clues, and the improved GM(1,1) model as a way to predict when thetarget block. Experiments show that the use of these methods can stable trace to thetarget under the change of illumination, the interference of analogue, and thecompletely blocked of target.In order to stably and quickly trace to the target, it is proposed based on particlefilter with improved the Least Squares Support Vector Regression (LSSVR). This methodis the use of an improved Kalman filter in the framework of particle filter to predict the target. And then it corrects the tracking results to reduce error arising from theprevious track through the LSSVR model. The Kalman filter tracking method ofreal-time is better than Camshift tracking method, so the particle filter with improvedLSSVR tracking algorithm is compared with the particle filter with improvedCamshift tracking algorithm can quickly track to the target.