Research On Visual Object Tracking Algorithm Via Kernelized Correlation Filters

Object tracking is one of the important research directions in the field of computer vision and artificial intelligence.It is of great worthy in the civil and military fields.Although significant progress has been made in object tracking,for the changeable of the object appearance,and the complex of the background in the actual tracking process,object tracking is still a very challenging problem.Therefore,This paper discusses the advantages and disadvantages of Kernelized Correlation Filter(KCF)trackers and conducts in depth research.The main innovative achievements are as follows:(1)In order to solve the problem of illumination variation in object tracking process,a new adaptive target tracking based on Kernelized Correlation Filter was proposed.Firstly,in order to improve illumination insensitivity tracking,color attribute was used to describe the target.Secondly,the local linear embedding algorithm was applied to reduce the dimension of extracted feature to achieve a low-dimensional feature space.Finally,the position was obtained by learning the regularized least-squares classifiers.The experiments demonstrate that the proposed algorithm not only has well illumination insensitivity,but also has higher tracking accuracy and robustness in the complex background.(2)In order to solve the problem of scale variation and occlusion in the object tracking process.An anti-occlusion object tracking algorithm based on Kernelized Correlation Filters is proposed.A multi-scale filter was introduced and scale estimation was obtained through calculating the maximum response value of the multi-scale filter.And according to the difference of peak sharpness of target position,then the model could be correctly updated.The experiments demonstrate that the proposed algorithm can not only effectively solve the target scale changes,complete occlusion and other issues in the complex background,but also has higher tracking robustness and accuracy,and the comprehensive performance has been significantly improved.(3)In order to solve the problem of fast motion and motion blur in the object tracking process,an adaptive target response based on Kernelized Correlation Filter was proposed.Firstly,with the KCF tracking framework,the boundary effect of the cyclic matrix was solved by add a prior target response to jointly optimize the training classifier.Then,to improve the robustness of the algorithm,the online support vector machine(SVM)classifier was applied to reposition the object to improve the robustness of the algorithm when the response value was less than the set threshold.The experiments demonstrate that the proposed algorithm can accurately and reliably track the object when the object is moving fast,blur and so on.In this paper,we focus on difficult problems of KCF tracker,and put forward the corresponding improvement algorithms.The experimental results show that the three algorithms were mentioned in this paper have good tracking performance in illumination variation,occlusion,scale variation and fast motion.