Research On Tracking Techniques Of Multiple Radar Emitter Targets Based On PHD Filter

After decades of research and development,the traditional multi-target tracking algorithms have been successfully applied in many engineering fields.But in the face of the more complex multi-target tracking scenarios,due to the complicated problem of data association the traditional tracking algorithms are difficult to deal with.However,the recently proposed tracking algorithm using the probability hypothesis density(PHD)filter which is based on the theory of the random finite set statistics,can give a strict and uniform mathematical description of the instances of target birth,spawning,disappearance,undetected measurement and clutter.It avoids the complicated problem of data association in the traditional tracking algorithms and has better tracking performance with lower computational complexity.Therefore,this dissertation focuses on the problems of multiple radar emitter targets tracking based on the PHD filter in the stages of tracking initialization,tracking maintenance and tracking termination respectively.The major contributions are given as follows:1.In the stage of tracking initialization,this paper mainly studies the problem of unknowing the target birth intensity in the PHD filter and proposes an adaptive estimation algorithm.According to the measurements from the neighboring time steps,both the single-point track initialization algorithm and two-point difference track initialization algorithm are used to calculate the positions and velocities of the candidate newborn targets.Then,the target birth intensity is estimated by the use of a Gaussian mixture form.Compared with the existing algorithms for the birth intensity estimation,the proposed method can represent the initial states of newborn targets more effectively,which strengthens the practicality of the PHD filter.On that basis,the prediction and update processes of the standard PHD filter are improved.After that,both the Gaussian mixture and sequential Monte Carlo implementations of the modified PHD filter are presented in the paper.2.In the stage of tracking maintenance,considering that the classification information of the radar emitter is different from but related to the classification information of target platform,a classification information aided Gaussian mixture PHD(GM-C-PHD)filter based on the signal features of radar emitter is proposed.This paper first makes use of the signal features to identify the radar types,then based on the transferable belief model the recognition results are transformed into the same frame of the known classification information according to the radar-platform affiliation.On the basis of that,their similarity measured by the compatibility ratio is used to approximate the likelihoods in the GM-C-PHD filter.As a result,the classification information can be used for radar emitter target tracking.3.Considering that the radar emitter recognition algorithm cannot provide accurate results in the complex electromagnetic environment,a novel GM-PHD filter using signal parameter information of radar emitter(GM-E-PHD)directly is proposed to track the radar emitter target.In the case of unknowing the distribution of signal features,inspired by the probabilistic data association(PDA),the membership degrees of the signal measurement belonging to the targets and clutter are calculated to approximate the likelihoods.Then,the emitter signal information can be integrated into the update process of Gaussian component weight of the GM-PHD filter and improves the multi-target tracking performance.4.In the stage of tracking maintenance,to address the problems of high computational complexity and inaccurate measurement set partitioning in the extended target PHD(ET-PHD)filter,an ET-GM-PHD filter based on the measurement set partitioning algorithm using the hierarchy clustering is proposed.For the given measurements,it combines the two most similar measurement cells to obtain new partitions step by step,and iteratively computes the likelihoods by the use of the neighboring partitioning results.As a result,the computational burden can be greatly reduced.Furthermore,considering the additional measurement information from the radar emitter target,the signal feature is also used in processes of partitioning the measurement set and updating the weights of the Gaussian components in the ET-GM-PHD filter to further improve the tracking performance.5.Considering that the existing ET-PHD filtering algorithms are difficult to model the extent of the group target accurately,a modified GM-PHD filter combining with clustering is proposed to track the group targets of radar emitter.In the update process of the GM-PHD filter,the proposed method firstly introduces the dummy measurements generated by the group centers to overcome the problem of undetected measurements.After estimating the target statements,they are clustered to achieve the group tracking.Then,the track points of the group centers from the neighboring time steps are connected to obtain the entire trajectories of the group targets.Simulation results show that the proposed method can effectively deal with the spawning and combination of the group targets,and provide a better tracking result of the group targets of radar emitter.6.Finally,in the stage of tracking termination,because there are many track segments causing the unsatisfactory tracking results in the emitter target tracking,combining with the labeling GM-PHD filter a multistage method based on the tracklet association is proposed to track the emitter targets.In the stage of tracklet generation,integrated with the adaptive estimation of target birth intensity and the emitter signal information,the labeling GM-PHD filter is applied to obtain the track segments of the radar emitter target.After that,in the stage of tracklet association,both the multipoint motion information and the emitter signal information are used to compute the similarities between the tracklets.The affinity propagation algorithm,which does not impose the constraint of one-to-one correspondence,is then adopted to cluster the tracklets.In the stage of association refining,the clustering result is adjusted to refine the final trajectories according to the spatial-temporal constraint of the tracklets.As a result,a more complete and accurate tracking result of the emitter targets can be obtained.