Copula application in nonlinear/non-Gaussian Bayesian tracking in the case of correlated sensors

Abstract

One of the most important challenges in target tracking is the modeling of correlated and non-Gaussian random processes. In this paper, a new target tracking approach by means of particle filtering in environments with highly correlated sensors, is discussed. The goal is to provide an accurate model of dependency structure in multivariate observation likelihood function, with non-Gaussian marginals to obtain a new algorithm in tracking problems. The main novelty of our method, termed as Copula-based Sequential Importance Resampling particle filter (CSIR-PF), is an application of the copula theory which is a powerful tool in correlation modeling in statistical theory. The precise model obtained by copulas makes a great improvement in particles' weightings. The performance of our proposed method is evaluated through the simulations of target tracking problems with highly correlated sensors. Results clearly indicate the acceptable performance of CSIR-PF.