Fast multiple target tracking using particle filters

Abstract

A fast algorithm for accurate initiation and tracking of multiple targets is developed. The basis of the approach is the linear multi-target (LM) method in which exact joint data association for multiple targets is approximated by single target data association for each target with clutter density modified to include contributions from neighbouring targets. This enables multiple target tracking to be performed with the computational expense of single target tracking. Previously the LM method has been combined with tracking using a Gaussian approximation. In this paper accurate tracking is achieved through the use of a variant of the auxiliary particle filter. Simulation results show that the use of the particle filter is particularly advantageous in demanding scenarios where targets are in close proximity for long periods.