Research On Object Tracking Technology Based On Correlation Filtering

Visual object tracking is an important research field of computer vision,which plays an important role in intelligent surveillance,security early warning,telemedicine,and military fields.The correlation filter is widely concerned by the industry because of its excellent tracking performance and high efficiency.Although the correlation filter has excellent tracking performance,it cannot perfectly track the target in the face of challenges such as rapid target movement,target deformation,illumination change and scale change in the practical application process.In this paper,the kernel correlation filter(KCF)is researched to improve its shortcomings.The main contributions of this article are as follows:(1)In order to solve the problem of insufficient detection range of the tracking algorithm when the object is moving fast and the loss of part of the target information caused by the use of cosine windows in cyclic sampling,this paper proposes a double template correlation filter(DTCF)based on KCF.In this algorithm,two scale-level filter templates are introduced.Firstly,the small template is used to predict the object position,and then the prediction information is judged to be reliable by the threshold value.If it is not reliable,based on the idea of secondary detection the filter uses a large template for secondary detection.The more reliable result of the two is selected as the final prediction position.The OTB-50 data set is used in the experiment,and the template-switching value is set to 0.6 according to the experiment.The experimental results showed that our algorithm was superior to the comparison algorithm in tracking performance.Compared with KCF,the mean values of DP,OP,S and CLE increased by 9.74%,14.81%,12.30% and 1.16%.(2)In order to solve the problem of feature robustness and object scaling when the tracking algorithm faces different challenges,this paper further proposes a multi-feature scale adaptive double template correlation filter(ADTCF)tracking algorithm based on DTCF.This filter trains a scale filter separately for object scale estimation according to the idea of a scale pyramid.At the same time,the shape,texture and color features are fused and applied to the translation filter and the scale filter to improve the tracking performance and robustness.The OTB-100 data set is used in the experiment.Compared with KCF,the experimental results showed that the mean values of DP,OP,S and CLE increased by 19.92%,38.78%,29.46% and 32.10%.