Research On Multi-detection Target Tracking Based On Random Finite Set

In recent decades,the technology of sensors has been continuously developed,and the range of applications for sensors is growing.Thus,a single target may produce multiple detections at one scan and standard target tracking algorithms are not applicable.It is urgent to propose algorithms that can be used for multi-detection target(MDT)tracking.Based on the theory of Random Finite Set(RFS),the following topics about the MD target tracking problem is studied in this thesis: implementation of RFS based multi-detection filter,measurement set partitioning algorithm for multi-detection filter,new filters that can be applied to arbitrary clutter and target measurement processes,enhancement of target detection performance by RFS method.The specific content is arranged as follows.Chapter 2 briefly introduces the theory of Finite Set Statistics(FISST)first.Then,several RFS based filters are introduced,including the Bernoulli filter for single target tracking,and the Probability Hypothesis Density(PHD)filter and Cardinalized PHD(CPHD)filter for multiple target tracking.Finally,the extended target PHD(ET-PHD)filter and the general PHD filter are introduced.Chapter 3 focuses on the multi-detection measurement model and the implementation of RFS based multi-detection tracking filter.Several multi-detection measurement models are introduced first.Then,starting from the multipath measurement model,the particle filter(PF)implementation of the point target multipath Bernoulli filter(MPBF)is proposed.Finally,a RFS based extended target multipath tracking algorithm and its PF implementation are proposed.The simulation results show the good tracking performance of the proposed algorithms in multipath scenarios.Chapter 4 presents a measurement set partitioning algorithm for multi-detection filters to reduce the computational complexity.Firstly,the RFS-based multi-detection filters,including the multi-detection Bernoulli filter and the multi-detection PHD(MD-PHD)filter,are introduced.Then,a generalized distance(GD)partitioning algorithm is proposed to reduce the number of subsets or partitions for the corrector of the filters.The simulation results show that the proposed partitioning algorithm can effectively reduce the computational complexity of the filters and maintain the tracking performance.Chapter 5 proposed 2 new filters that can be applied to arbitrary clutter and target measurement processes.The one is the general CPHD filter,and the other is the general Bernoulli filter,the single target version of the general CPHD filter.After that,it is proved that the general CPHD filter is the theoretical basis for various types of CPHD filters,PHD filters and Bernoulli filters.In order to reduce the computational complexity of the proposed general CPHD filter,a general partitioning algorithm is proposed.According to the simulation results,compared with the MD-PHD filter,the variance of the target number of the general CPHD filter is smaller,and the state estimation is more accurate.Chapter 6 proposes a RFS based data level moving target detection algorithm.Firstly,whether the target exists is modeled as whether the target track exists.Then,on basis of the Bernoulli random finite set(B-RFS),the data level moving target detection algorithm and its importance sampling implementation are proposed.The simulation results show the good detection performance of the algorithm.This chapter demonstrates that the RFS method can enhance the detection performance.Chapter 7 is a summary of the full text and the prospects for the further works.