Research On Video Target Tracking Algorithm Based On Kernel Correlation Filter

Video target tracking has very important value and broad application prospects in man-machine interactive,video surveillance and medical diagnosis.However,due to the diversity of the characteristics of the target and the complexity of the environment,the tracking algorithm KCF(Kernelized Correlation Filters)can not be used to obtain effective and accurate tracking.The key factor to limit the robustness of the tracking algorithm is appearing in the process of tracking the target scale changes and occlusion problems.So for the shortcomings of KCF algorithm,this paper mainly optimizes and implements the algorithm.The main work of this paper has the following aspects:(1)KCF algorithm for the problem can not cope with the target scale changes,a new optimization algorithm: the scale is detected by the target size estimation of ways to further determine the optimal scale frame target to be tracked in the current.Firstly,by analyzing some current target scale algorithms,the target scale transformation scheme is determined.By introducing the DSST(Discriminatiive Scale Space Tracker)algorithm to improve the MOSSE(minimum output sum of squared error)algorithm,and using the bilinear interpolation method to obtain the best scale,the problem of unable to cope with the target scale transformation is solved.Based on this idea,this paper improves the DSST algorithm and incorporates it into the KCF algorithm,and optimizes the KCF algorithm.(2)KCF algorithm for tracking occlusion caused by loss problems,a new optimization algorithm: Adding occlusion detection mechanism to determine if the target is blocked,then reduce the model update factor,thereby reducing errors introduced due to the occlusion information.First,the detection scheme is determined by analyzing some current detection algorithms.Forward and reverse tracking is performed on the target timing,and the trajectory error of the forward tracking and the backward tracking is compared to determine whether the target has occlusion.If the occlusion detection judgment mechanism is activated,the template updating method is modified at the same time.(3)This paper also discusses the follow-up evaluation methods,the application of the principles introduced an improved method of KCF.Experiments were selected and the presence of the target scale changes plurality of sets of video presence of occlusion,the improved algorithm is used to test these videos respectively.Compared with the original KCF algorithm qualitative and quantitative experiments,it is seen from experiments that the improved method can be well applied to track changes in scale change and occlusion.