Distributed Multi-sensor Multi-target Tracking Inthe Framework Of Random Finite Sets

In this thesis,for the purpose of monitoring an unknown number and time-varying number of targets in a distributed sensor network,we investigate the work of distributed multi-sensor multi-target tracking(DMMT)technique.Compared with centralized multi-object tracking solutions,the DMMT methods generally benefit from lower communication cost and higher fault tolerance,hence the theoretical research of the DMMT and its technical applications are of great significance.As an emerging technique,the DMMT has its own problems and challenges.To this end,the work about distributed fusion algorithms design,the efficient implementation of distributed fusion,the network consensus technology of distributed fusion are studied in the framework of random finite set(RFS),which provide a technical way for DMMT.The main contributions of the thesis are as follows:1.In order to tracking multiple targets,this thesis modelled the multi-target tracking problem for single sensor using multi-Bernoulli(MB)filter in the framework of RFS.For the unknown correlation among different sensor nodes,this thesis proposed a distributed multi-object tracking algorithm through the use of the MB filter based on Generalized Covariance Intersection(GCI),referred as GCI-MB fusion.This algorithm derived an approximated tractable expression of GCI fusion with MB distributions,which provides a precondition for the fusion algorithm.Compared with GCI-PHD and GCI-CPHD filters,the GCI-MB fusion can be more efficient and accurate in problems that require particle implementations or target individual existence probabilities.2.The original GCI-MB fusion involves the computation burden which grows exponentially with the number of targets,for the purpose efficient implementation,this thesis proposed a parallel GCI-MB grouping(GCI-MB-G)fusion.This algorithm derived a fully parallel fusion structure,under which,the problem of distributed high-dimensional multi-target fusion turns into multiple distributed multi-target fusion in low-dimensional.Our analyses show that the proposed algorithm can reduce both the computational expense and memory dramatically with slight approximation error.This efficient solution can serve as the technical support for the application of GCI-MB fusion in practical scenarios.3.In order to average all the multi-target posteriors from the nodes in a distributed sensor network,this thesis proposed the consensus based distributed multi-object tracking with MB,referred as C-GCI-MB fusion.By iteratively using information communication among sensor nodes,and then local fusion strategy,we implement the C-GCI-MB fusion.The C-GCI-MB fusion can avoid the double counting phenomenon which is easy to arise when utilizing consensus technology.The proposed algorithm enables the distributed sensor network share information among sensor nodes,distributed computing and global estimation of target states.