Distributed Particle Filter For Target Tracking In Networked Radar Systems

Networked radar system(NRS)is a particular sensor system that composed of multiple network connected radars,which is capable to expand the spatial and temporal coverage,improve the detection ability of the target and increase the robustness reliability of the system.Therefore,the NRS is widely applied to numerous fields,such as target detection,target localization,target tracking and area surveillance.In these applications,target tracking is one of the most important research topics.In particular,the distributed particle filtering(DPF)algorithm for target tracking based on the optimal Bayesian rule,is studied.In this paper,we focus on some important issues of the NRS,for example,parameterized the local information,fusion of the correlation information and non-synchronous information fusion in NRS,etc.Generally,this paper is consisted of the following three parts.1.In this part,we first give a brief summarize of the system architectures and models of the NRS,then some typical target tracking algorithms based on the optimal Bayesian estimation rule are introduced.2.To address the two main issues in synchronous NRS,i.e.,parameterized the local information and fusion of the correlation information,two distributed target tracking algorithms are proposed,respectively.Herein,for the polynomial approximate likelihood based DPF algorithm,the communication cost is greatly saved since the likelihood function is represented by polynomial parameters.For the importance sampling posterior based DPF,the local posterior probability density function(pdf)is approximated by a set of weighted samples.Then the particle samples are fused in an exponential weighting form based the Chernoff fusion rule,which efficiently addressed the problem that fusion of the correlation information.In the simulation,the efficiency of these two algorithms are shown,and some important parameters are analyzed.3.To address the problem that non-synchronous information fusion in NRS,two DPF algorithms for target tracking in non-synchronous NRS are proposed,where we named as likelihood-based DPF for non-synchronous NRS and posterior-based DPF for non-synchronous NRS in this paper.The basic idea of these two algorithms are establishing the efficient non-synchronous measurement model according to the specified NRS at first.Then,by incorporating a time-aligned strategy,the asynchronous measurements among the update period are fused to jointly estimate the target states.The simulation results show the main advantages of the two proposed algorithms,such as low computation cost and robustness,and some key parameters are analyzed.