Target Tracking Algorithm Based On Correlation Filtering And Target Redetection

With the development of computer vision,the requirement of target tracking becomes higher and higher in our lifes,so the requirement of the target tracking algorithms becomes more sophisticated.The existing target tracking algorithm can solve target-tracking in most simple cases,but there are still lots of targets in complex cases can't accurately tracking.So there are still many new target tracking algorithm pulished every year.Target tracking algorithm based on correlation filtering was presented according to the principle of correlation,it has the characteristics of high efficiency and robustness,it is a kind of excellent target tracking algorithm.But the algorithm still has some problems to be improved,such as the determination of the search domain,the determination of feature extraction method,the loss of the target of deformation or fast moving,and the loss of the target because of obstructions.In this paper,the deficiencies of the target tracking algorithm based on correlated filtering is the point,in particular,Circulant Structure of Tracking-by-detection algorithm is selected to be improved with determination of the search domain,the determination of feature extraction method,and the loss of the target of deformation or fast moving.The main work and innovation of this paper are as follows:(1)Firstly,analyze the target tracking in the field of computer vision,describe the current situation and the problem that still need to be sloved in detail.Then introduce the idea and principle of target tracking algorithm based on correlation filtering,describe its specific process and several related algorithms,and propose the deficiencies of target tracking algorithm based on correlation filtering.Secondly,introduce the concept of image feature extraction,list several common feature extraction methods.Thirdly,deduce the formula of the CSK algorithm in detail,expound the core idea and the concrete steps of the algorithm.(2)This paper mainly aims at improving the CSK algorithm.In order to better train the filter,extract better featurea to represent the target,use the property of entropy to have adaptive selection of search domain size.For each frame,you can get a background area that fits the target size,this method can improve the accuracy of the algorithm.The traditional correlation filters target tracking algorithm usually only extracts single feature.In order to make the representation of the tracking target more reliable,this paper use the parallel extraction method of HOG feature and CN feature.After the correlation filter there are two outputs,choose better one as the final output,this helps the algorithm still be a highly robust tracking in many complex environments.For a constant in the algorithm,in order to make the matrix in the calculation closer to the Gaussian matrix,the constant is improved to a Gaussian distribution to improve calculation accuracy.The traditional target tracking algorithm often shows loss of target.In order to reduce th occurrence of this kind of situaton,the concept of redetection is proposed,the target center point of each frame is measured.If the center point offset is detected,reposition it and get a more accurate target center point,this method improves the accuracy of target tracking greatly.(3)For all the improvements in this paper,the algorithm is tested and validated on the dataset and compared with other algorithms,the results show the effectiveness of the algorithm.