Research On Multi-target Tracking Method With Unknown Parameters

In recent years,random finite set(RFS)has become a hot topic in the field of multi-target tracking because of its ability to avoid data association.However,the multi-target tracking filtering algorithm based on RFS generally has the problem of mismatching of target motion state parameters and inaccurate estimation of target state when the speed of the tracked target is unknown or the preset speed is greatly different from the real speed of the target.Therefore,this paper focuses on the unknown target velocity parameters of several filtering algorithms under the RFS framework,and proposes unknown estimation and calibration methods of relevant velocity parameters.The specific research contents are as follows:Firstly,the time transfer of multi-objective filter in the form of RFS through the optimal Bayesian filtering framework is studied.In order to reduce the algorithm complexity,a suboptimal approximate filtering algorithm,namely probability hypothesis density(PHD)filter,is proposed.Several classical multi-target filtering algorithms and common multi-target tracking performance evaluation methods are summarized.Secondly,aiming at the problem that the target state estimation performance of GM-PHD filter is poor when the target velocity is unknown or inaccurate,a motion parameter estimation combined smoothing filtering algorithm based on GM-PHD is proposed.In this algorithm,the speed information is extracted from target states,and the speed estimation accuracy is improved through median smoothing and linear smoothing,and then the speed is fed back to the state transition equation of GM-PHD filter to improve the prediction accuracy.Simulation results show that the proposed algorithm can significantly improve the state estimation performance of GM-PHD filter when the target velocity is unknown or inaccurate.Finally,GM-PHD filter does not involve target problem,only to determine a single target tracking,target quantity and intensive scenario to detect and track multiple targets at the same time speed parameters estimation problem,put forward a kind of based on generalized label multi-Bernoulli(GLMB)the unknown parameter estimation algorithm of filter.Each target should be marked and distinguished thus their state,position and speed information corresponding.With the tag variable added to the target state to associate the state information of the same target at different time,the state information of each target is obtained,and the speed information accuracy of each target is improved by smoothing algorithm,so that the multi-target Markov density can reflect the multi-target motion model more truly.The simulation results show that the proposed algorithm can achieve good tracking performance for both linear and nonlinear scenes under the condition of unknown speed,and the performance is better than the improved GLMB filter.