Auxiliary unscented particle cardinalized probability hypothesis density

Abstract

The probability hypothesis density (PHD) filter has been recently introduced by Mahler as a relief for the intractable computation of the optimal Bayesian multi-target filtering. It propagates the posterior intensity of the random finite set (RFS) of targets in time. Despite serving as a powerful decluttering algorithm, PHD filter still has the problem of large variance of the estimated expected number of targets. The cardinalized PHD (CPHD) filter overcomes this problem through jointly propagating the posterior intensity and the posterior cardinality distribution. Unfortunately, the particle filter implementation of the CPHD filter suffers from lack of an efficient method for boosting its efficiency other than the inefficient Bootstrap particle filter. We propose auxiliary unscented particle implementation of the CPHD filter as a solution to this problem. Numerical simulations indicate significant improvement in the estimation accuracy of the proposed algorithm over the available Sequential Monte Carlo (SMC) implementation of the CPHD filter.