A t-distribution based particle filter for target tracking

Abstract

From a Bayesian perspective, target tracking is the problem of generating an inference engine on the state of a target using a sequence of observations in time, which is to recursively estimate the probability density function (PDF) of the target state. Previous approaches to density estimation have mostly focused on Gaussian filters in practice, but these are well known sensitive to outliers. In this paper, the new particle filter is developed based on Student-t distributions, which are heavier tailed than Gaussians and hence more robust, called the Student-t distribution particle filter (SPF). To estimate PDF of the target state based on samples, a new expectation conditional maximization either (ECME) algorithm is developed and embedded in the SPF. The new ECME algorithm has a faster convergence rate than that of the existing EM algorithms. Under the Student-t distribution assumption, it has been shown that the Student-t distribution particle filter is asymptotically optimal in terms of the number of particles. Simulations have demonstrated the effectiveness and the improved performance of the SPF over Gaussian filters and the bootstrap filter (SIR), and have shown that it is more robust than SIR.