Research On Target Tracking Algorithm Based On Improved Kernel Correlation Filter

With the rapid development of artificial intelligence,target tracking technology has been widely used,such as automotive autonomous driving,intelligent monitoring,human-computer interaction,intelligent transportation and other fields.Due to its wide application fields,the application scenarios have become more complicated.At present,target tracking technology is plagued by many factors such as scale changes,complex background,partial or complete occlusion,rotation,and light changes.These factors have become difficult problems in the tracking field.In recent years,the target tracking algorithm has achieved fruitful results.Among them,the target tracking algorithm based on Kernel Correlation Filter(KCF)is the most successful.This paper mainly conducts related research on some problems existing in the scene such as target scale change and occlusion through the KCF algorithm to improve the tracking performance of the tracking algorithm in complex scenes.The main research work of this article is as follows:(1)A KCF target tracking algorithm based on feature fusion is proposed.For the original KCF algorithm,a single histogram of oriented gradient(Histogram of Oriented Gradient,HOG)feature is used.When the target occlusion,scale change,rotation,light change,etc.cannot fully describe the target appearance information,the algorithm has tracking accuracy and robustness.For the problem of poor sex,this paper first adds the Color Names(CN)feature that can distinguish different color information of the target when collecting basic samples;secondly,extracts the HOG and CN features separately,and extracts the extracted HOG and CN features Fusion is done in a weighted way,and the fused features are used as the target's appearance description.Finally,the nature of the cyclic matrix is used to perform cyclic shift operations on the basic samples to construct a large number of sample sets,which are obtained through the ridge regression model and kernel function training sample set The kernel correlation filter uses the kernel correlation filter to detect the target,and the position of the target in the image is determined by the output maximum response value.The research results show that the proposed feature fusion algorithm has better anti-interference ability in complex scenes than the original KCF algorithm,and improves the tracking accuracy of the algorithm.(2)A KCF target tracking algorithm against occlusion interference is proposed.In view of the problem that the original KCF algorithm is prone to tracking loss when the target is occluded,an occlusion determination model is added in this paper.The average value of the peak sidelobe ratio of the previous frame and the current frame in the image is used as the threshold for determining whether the target is occluded.If the peak sidelobe ratio of the frame is less than the current frame threshold,it is determined that the target is blocked,and then the URTS-EM(Unscented Rauch Tung Striebel-Expectation Maximum)algorithm is used to predict the target's trajectory,that is,the blocked target,And then feed back the predicted target position to the original KCF algorithm.The research results show that this method has good robustness when the target is blocked.(3)A KCF target tracking algorithm with adaptive scale change is proposed.Aiming at the problem that the tracking frame of the original KCF algorithm is not small and cannot be adapted to the target scale change in real time,this paper adds a step of estimating the target scale under the framework of the original KCF,and uses the kernel correlation filter to obtain the target position At this point,a plurality of samples of different scales are collected to train a one-dimensional scale filter,and the target is detected by the scale filter,and the size of the target scale is estimated according to the detected maximum output response value.The research results show that the tracking frame of the algorithm in this paper can change adaptively with the change of target scale,which greatly improves the real-time performance of the tracking algorithm.The content of this article is mainly based on the relevant research of the current popular KCF target tracking algorithm,and gives a solution,that is,the algorithm proposed in this article.In the experiment,the classic tracking algorithm and the test video sequence compared with the improved KCF algorithm proposed in this paper are from OTB(Object Tracking Benchmark),and the performance of the tracking algorithm is evaluated by the OPE(One-Pass Evaluation)evaluation method.Experimental research shows that the improved KCF algorithm proposed in this paper is effective and feasible,solves the problem of occlusion and scale change of the target,and improves the tracking accuracy and robustness of the tracking algorithm.