Research On The Algorithm Of Target Tracking Based On Feature Points

With the rapid development of information technology, the target tracking technique based on thevideo has a wide application value in many fields.We can achieve target tracking easily through thehuman eyes or a large number of prior knowledge under complicated background. But how to obtainrelevant information of the moving target through the streaming video to realize the target tracking is achallenging job.The principle of target tracking algorithm based on feature points was researched. The algorithmwas divided into feature point extraction algorithm and feature point matching two parts.In order toadapt the instantaneity of the target tracking system, FAST feature detection algorithm was chosedand improved by homomorphic filtering and adaptive threshold value which can adapt to the situationof different illumination and contrast to a certain degree and achieve the goal of suppressing multiplefeature point blocks. During the process of feature points matching, the pyramid KLT optical flowmatching technology was used. To solve the problem of unstable feature points, the method“matching before and after” was proposed to remove unstable feature points,so that the system ismore accurate.Owing to the continuously rejecting of unstable points, changes of target’s pose andenvironmental disturbance during the tracking process, the number of effective feature points of thetarget is gradually reduced and even becomes zero. To solve this problem, the paper put forward theway of combining the feature point tracking with particle filter. And to solve the problem of samecolor interference, the covariance matrix technology was used to describe the target. The experimentalresults showed that particle filter algorithm based on covariance matrix is capable of resisting thesame color interference to a certain degree and this algorithm is very stable.A kind of target tracking strategy was proposed to fit the situation where the target is occluded.With the recovery of occlusion, a new recognition of moving target was achieved through constantlyexpanding the search area, fusing target’s gradient, gray,color and other features. The experimentalresults showed that it can identify the target accurately in the situation of short occlusion.