A t-Distribution-Based Particle Filter for Target Tracking

For the predictive distribution and posterior distribution problem of target states in Bayesian tracking,a new particle filter,called the Student-t distribution Particle Filter(SPF),is developed based on multivariate student-t distributions by using sequential Monte Carlo methodology.To estimate probability density function(PDF) of the target state based on samples,a new Expectation Conditional Maximization Either(ECME) algorithm is introduced and embedded in the SPF.Under the student-t distribution assumption,it is shown theoretically that the SPF is asymptotically optimal in terms of the number of particles.In the tracking of maneuvering target,the performances of SPF,unscented Kalman filter(UKF) and the bootstrap particle filter(SIR) are compared in terms of accuracy.