Research On Robust Visual Target Tracking Algorithm Based On Particle Filter

Tracking and analysis moving target in the video is one of the key research in thefield of computer vision, because it has a wide range of applications in intelligenttransportation, human-computer interaction, and machine vision field. Particlefilter-based video object tracking is one of the most important method of trackingmoving target in the video, because it can do well in nonlinear and non-Gaussianproblem and it also gets better effect in tracking the occlusion object in the video. Inthe current, it is a research hotspot in the video-based moving target tracking.So far, particle filter technology for video tracking faces two major problems.First, the target itself in the process of tracking changes a lot, and background andlight have a bad influence on the characteristic of the target. Second, the target can beoccluded by some similar objects. For the problem of tracking drift brought by thetarget apparent change, we adopt the way of multiple cues fusion so that our targetapparent can be more in line with the objectives of the real situation. Through theanalysis of different features, we combine color features, texture features, edgefeatures and motion features as target apparent characteristic. Experiments show thatthe particle filter algorithm based on multi-feature fusion has a higher accuracy androbustness. For the problem of occlusion, the target feature changes a lot. So, weconsider the sub-block processing of the object. By doing this, we can identify thetarget from the information of sub-block which is not being obscured. Experimentsshow that the method has better accuracy.In addition, when the target gets easy to lose in the case of tracking target movesfast, we can give the particle a little offset which is consistent with the direction ofmovement according to the historical movement of the target. It achieves good results.