Research On Multi-target Tracking Method Based On Multi-feature Matching

Today,radar target tracking technique has been widely used in military and civilian fields.As the scene of radars to surveil gets more and more complicated,many factors including noise and clutter severely interfere performance of multi-target tracking methods.Most of traditional target tracking algorithms use only the position information of the measurement point to track targets,which makes it easier for a target track to be associated with a wrong measurement point.As a result,the precision of the target tracking algorithms fails to meet the demand of practical applications.Aiming at this problem,this thesis investigates multitarget tracking methods using multi-feature matching.By using the position information and other feature information in measurement points,the proposed method improves the precision of the measurement point association and the precision and stability of tracks of multiple targets in a complex scenario.Moreover,simulated data and measured radar data are used to verify the effectiveness of the proposed methods.The content of the thesis is organized as follows:The first chapter introduces the research background of the thesis as well as the significance,history and current situation of radar target tracking.The content of the thesis is summarized.The second chapter firstly introduces the fundamental knowledge in radar target tracking.The general process of radar target tracking is briefly reviewed.The concept of the correlation window,widely used in radar target tracking,is introduced.One of the classic algorithms,Kalman filter and derivative algorithms form it are introduced.In the third chapter,traditional algorithms of track initiation are introduced.Aiming at the fact that traditional algorithms of track initiation are easy to generate multiple suspected initial tracks in a complex scenario,the value of the test statistic and the estimate of Doppler frequency in target detection are introduced to improve the quality of the track initiation and a method to evaluate initial tracks using multiple features is given.At last,experiments using measured radar data are used to show the fact that the proposed method can improve the efficiency and accuracy of track initiation.The fourth chapter investigates application of multi-feature matching in point association.First of all,the three common point association methods are reviewed.Next,in order to solve the problem of low accuracy of these association methods in a complex scenario,a point association method based on multi-feature matching is proposed.Moreover,it is shown by simulation that under different clutter density environments the improved association method can obtain better performance than the traditional methods in both single target tracking and multi-target tracking scenarios.In the fifth chapter,the multi-feature matching is utilized to improve Gaussian mixed probability hypothesis density(GM-PHD)filter.Firstly,the theory of random finite set and the principle of GM-PHD filter are reviewed.Moreover,aiming at the problem of low precision of the weight calculation of GM-PHD filter in the iteration of Gaussian component,a GM-PHD filter using multi-feature matching is presented.At last,it is shown by simulations that the improved GM-PHD filter can adapt to scenarios such as target track crossing,target birth and target termination more effectively and gives more accurate estimates of target number and location.The sixth chapter summarizes the work of this thesis and discusses the further research.