Research On Moving Target Tracking Method Based On Correlation Filter

Recent years,with the development of Internet technology,motion target tracking technology has been widely used in many aspects of society.However,this kind of technology is not yet fully matured.For example,in some complex sports environments,when the target undergoes deformation,motion blur,or occlusion changes,general algorithm can not deal with the sudden changes of two or more targets in time,and prone to drift,misalignment etc,and finally lead to a failure of tracking.This paper studies and analyzes some existing mainstream tracking algorithms,proposed an improve method and verifies its robustness.The tracking algorithm based on the correlation filter is a mainstream branch in the current tracking field.It uses a feature of target to train a filter and transforms the complex calculations in the time domain into dot multiplication between elements in the Fourier domain.Candidate regions are correlated with the filter,and the target is located at the place where the response value of the filter output is maximum to the corresponding area.Since the tracking algorithm based on the correlation filter improves the tracking speed,and it's accuracy is relatively good,this paper studies on this basis and starts work from the following aspects:(1)First of all,the classical tracking algorithm Calman filter,particle filter and correlation filter tracking algorithm are analyzed and compared to determine the framework of the correlation filter.Then the general tracking process based on the correlation filter is introduced in detail,including the feature extraction of the target,the training filter and the update filter,and the different methods used by different algorithms in each process are briefly introduced.(2)The traditional target tracking algorithm based on Color-Naming process that describe the target with a novelty color feature,which has the advantages of less calculation and good real-time performance.However,the traditional Color-Naming algorithm does not perform well when the target is scaled.So we introduce an image scale pyramid to improve it,design an adaptive target tracking algorithm based on color features,and use the benchmark sequences to verify our method for quantitative and qualitative analysis.(3)In a real-time tracking video,the target often moves in a complex environment,only using the color feature to describe may confuse with the background when the color is enormously similar with each other.We introduce HOG(Histogram of Oriented Gradient),which is a local feature,to describe the target combining global features with local features,and it would be more comprehensive.Experiments show that the algorithm proposed in this paper can not only track the target accurately but also overcome the shortcomings of color features and HOG features under complex motion scenarios,such as camera shake,unrecoverable deformation of target appearance and similar color of target and background,which has better robustness.