Application Of Particle Filtering To Multiple Target Tracking

Target tracking is an important element of surveillance, guidance, or obstacle avoidance system, whoes role is to determine the number, position, movement, and identity of targets. The fundamental building block of a tracking system is a filter for recursive target state esimation. Kalman filter is the best known filter, a simple and elegant algorithm formulated more than 40 years ago, as an optimal recursive Bayesian estimator for a somewaht restricted class of linear Gaussian problems. Recently there has been a surge of interest in nonlinear and non-Gaussian filtering. A number of techniques for this type of filtering are reviewed in this thesis, but the main focus are the tools of sequential Monte Carlo estimation, collectively referred to as particle filter.In chapter one, we reviewed the history, current situation and application areas of informatin fusion, target tracking and estimation theory(filtering).Chapter two give a survey of the common filtering algorithms for linear, Gaussian problems and for non-linear Gaussian problems. The Bayesian estimation framework is also introduced as the fundation of the recursive filtering algorithm.In Chapter three, we first introduce the Monte Carlo method for Integral. Then importance density and resampling are discussed. At last all these techniques are blended into the SIS framework(the generic particle filter framework). This chapter also discussed several variants of the particle filter in SIS framework.In charpter four, we use EKF, UKF, SIR to tracking a single target in typical nonlinear environment. The result show the SIR algorithm has its superiority over the other two methods. The application of particle filter solving joint multiple target tracking is also addressed in this chapter.Finally, chapter five summarizes the main benefits, challenges and its future of particle filter and its application for target tracking.