Research On Target Tracking Algorithm Based On The Kernelized Correlation Filters

Target tracking has a wide application prospect in human-computer interaction,intelligent monitoring,autonomous navigation,military defense and so on.However,due to the problems of occlusion,deformation,illumination,background clutters,motion blur during tracking,the target tracking technology still faces many challenges in application.The target tracking algorithm based on kernel correlation filter(KCF)has received wide attention due to its fast speed and high accuracy,but it still exists tracking failure.In order to further improve the percision and success rate of the KCF algorithm,this paper makes an deep study of it,and the following improved algorithms are proposed :(1)In order to describe the target better,the feature fusion is proposed in this paper.The complementary HOG features and CN features are integrated together,giving full play to the advantages of each feature and enhancing the ability to characterize the target.(2)In order to solve the problem of the lack of scale estimation in KCF algorithm,a method of scale estimation is given in this paper.First,a series of image blocks with different scales are collected around the center,and training a one-dimensional scale correlation filter.Then the filter is used to detect the target,and the optimal scale of the target is predicted.(3)Aiming at the problem of tracking failure of KCF algorithm under occlusion,deformation and motion blur,this paper proposes an improved algorithm based on multi peak re-checking strategy.This algorithm uses the filter response graph to recheck the target,and avoids the tracking failure caused by the error locating target.In order to verify the effectiveness of the improved algorithm in this paper,73 groups of color video sequences in the OTB-2015 data set are tested in the experimental section.The results show that the precision and success rate of the proposed algorithm are 77% and 56.7%,which are obviously higher than the existing popular algorithms.At the same time,experiments show that even in complex conditions,the algorithm can still track targets in a stable and accurate way.