Research Of CT Object Tracking Algorithm In Complex Environment

Object tracking has a wide range of applications in many fields,such as human-computer interaction,intelligent surveillance,virtual reality and so on.It is one of the hot topics in the field of computer vision.Compressive tracking(CT)is a simple and efficient object tracking algorithm.Once proposed,it had received much attention.However,CT algorithm may drift or loss the target because of the complexity of the tracking environment in reality.In order to improve the performance of CT algorithm and make it robust in complex environment,this paper makes a basic research on CT algorithm and proposes some improved algorithms based on it.The works are as follows:(1)Since CT algorithm is based on compressed sensing theory,this paper introduces much knowledge about it,and expounds three stages in compressed sensing processing respectively.And the basic framework and the principle of CT algorithm are introduced.What's more,some of the deficiencies and improvements of CT algorithm were analyzed.(2)Fast compressive tracking combined with Kalman filter is proposed.CT algorithm may drift or loss the target when the target is occluded.In order to solve this problem,the Kalman filter is applied to CT algorithm.When the target is occluded,the Kalman filter is used to track the target.And the parameters of the Kalman filter are not updated anymore until the target reappears.If the target is not occluded,CT algorithm is used for tracking,and the information tracked by the CT algorithm is used as the input of the Kalman filter.The predicted position is taken as the next initial search location of CT algorithm to find the target location faster.CT algorithm uses a fixed learning rate to update the classifier model,which will lead to the accumulation of errors.And it may result in tracking drift.In order to solve this problem,the adaptive learning rate is used.The experimental results show that the proposed algorithm has a better tracking performance than CT algorithm and other traditional tracking algorithms when the target is occluded.(3)Weighted compressive tracking algorithm with adaptive scales is proposed.The target scale cannot update adaptively in CT algorithm.In order to solve this problem,the adaptive scales is used in CT algorithm.Firstly,the features at the different scales track windows are extracted.The scale pyramid is composed of these features,and then it isinput as a feature of the correlation filter score.Finally,The new scale was then found by maximizing the score.The Bayesian classifier performance used in CT algorithm is not strong.In order to improve its performance,Bayesian classifier is weighted and the similarity of positive and negative samples is taken as the weight.The similarity between the two is smaller and the weight is higher.Experimental results show that the proposed algorithm has a better tracking performance than the CT algorithm and other traditional tracking algorithms when tracking the target whose scale changes greatly.