Research Of Sequential Bayesian Filter In Multi-target Tracking

The major purpose of multiple target tracking is to detect individuals target and estimate their states in the presence of clutter and uncertainties of measurement and motion mode.The traditional multi-target tracking algorithm usually uses data association technique,but there may be several problems in data association technique,such as “combinatorial explosion” and the exponential growth of computation load.Recently,the PHD filter algorithm based on Finite Set Statistics(FISST)was proposed by Mahler.It not only avoids data association,but also solves the problem of multi-target tracking in the presence of false alarms,missed detections and an unknown number of targets.Although the PHD filter has many advantages in multi-target tracking,it has several problems that need solving.Firstly,the PHD filter cannot distinguish multiple different targets when they are closely spaced.Secondly,the PHD filter handles received measurements periodically,which leads to a delay in data processing.Finally,in case of low detection probability,it is prone to resulting in the loss of the target information and the instability of the target number estimation.Aiming at these problems of the PHD filter,we propose a sequential multi-target Bayesian filter.In order to make the filter accommodate multiple maneuvering targets tracking,we propose a sequential multiple target Bayes filter with jump Markov system(JMS)models to track multiple maneuvering targets.The main content of this thesis can be summarized as follows:1)In Chapter 2,we introduce the theory of the multi-target Bayesian filter based on FISST,discuss the model of multi-target tracking,and summarize the optimal multi-target Bayesian filter and the PHD filter that propagates the first-order moment of the joint distribution.Finally,we analyze the multi-target Bayesian filter for propagating marginal distribution and existence probability.2)In Chapter3,based on our research on multi-target Bayesian filter,we propose a sequential multi-target Bayesian filter.This filter propagates marginal distributions and existence probabilities of each target,and sequentially handles the measurement data received at the current moment.In addition,we also present two implementations of sequential multi-target Bayesian filter to accommodate linear and nonlinear Gaussian multi-target models,respectively.The simulation results demonstrate that the proposed filter can track multi-target better in the presence of clutter,miss-detection and an unknown number of targets than the PHD filter.3)In Chapter 4,in order to solve the problem of multiple maneuvering target tracking,we combine the sequential multi-target Bayesian filter with JMS approach to develop the sequential multi-target Bayesian filter for JMS models.We also present two implementations of sequential multi-target Bayesian filter for JMS models to accommodate linear and nonlinear Gaussian multi-target models,respectively.The simulation results demonstrate that the proposed implementations are more efficient at multiple maneuvering target tracking in the presence of clutter and uncertainties of measurement and motion mode than the PHD filter for JMS models.