Research On The Technology For Multi-target Tracking Based On Random Finite Set

Multi-target tracking technology is currently one of the highly attention of theresearchers，it has important application value and broad development prospects. With thehigh-technique developing quickly, more and more complex requirements of the theoreticdevelopment have been set forth by various applied systems, the traditional theory andtechnique of multi-target tracking can not face the new challenges. In recent years, theapproaches to MTT based on random finite set (RFS) have attracted more attentions. Somescholars have proposed a variety of excellent filtering algorithm，The PHD filter and CPHDfilter are two kinds of the most representative algorithms.Based on the RFS of multipletarget tracking algorithm can effectively solve the difficulties of traditional multiple targettracking method. So the study on multiple targets tracking technology based on RFS theory isof academic and realistic significance.This article is mainly do research around the multiple target tracking technology basedon RFS. Based on the in-depth understanding of the PHD filter, we give the BFG-based GM-PHD smoothing filter, ET-PHD filter and GIW-PHD filter. The specific research contents areas follows:Research on multiple maneuvering targets tracking algorithm based on the GM-PHDfilter.A jump linear or nonlinear gaussian markov model is established and a BFG-basedGM-PHD filter is proposed.New algorithm adopts a simple best-fitting Gaussianapproximation method to express the model dynamics，and uses the forward-backwardsmoothing smoothing the PHD filtering result.In addition, new algorithm uses a secondaryconsolidation and extraction of state method, which in order to obtain more accurate targetnumber and state estimation.Research on multiple extended targets tracking algorithm based on the ET-GM-PHDfilter. The computation problem in the process of implementation is discussed.Then a distanceheuristic segmentation method and a partition strategy based on K means clustering methodare presented, which can effectively reduce the amount of calculation of the algorithm. A newform of ET-GM-PHD filter is put forward.This algorithm allows part of the target birth model to take on a uniform distribution, thus obviating the need to use large Gaussian mixtures priorparameters to approximate a birth density.Research on multiple group targets tracking algorithm based on the GIW-PHD filter. Forthe computation problem of the GIW-PHD filter in the implementation process, a new form ofdistance heuristic segmentation method and a merging algorithm based on Kullback-Leiblerdivergence for reduction of Gaussian inverse Wishart mixtures are presented. Made areasonable improvement to the standard GIW-PHD filter, and develop a new form of GIW-PHD filter. The new approach can model extension dynamics and measurements moregenerally and reasonable, it also can obtain a good estimate of the Measurement Rate of time-varying.