Particle filters and data association for multi-target tracking

This paper presents Monte Carlo (MC) methods for multi-target tracking and data association. We focus on comparing different estimation methods based on joint and non-joint state particle filters (PF) and joint probabilistic data association (JPDA) techniques. A novel data association algorithm for PF, founded on a combination of PDA and nearest neighbour (NN) techniques, is also developed. In this method the calculation of the association probabilities for each target is simplified and especially in clutter environment the number of association hypotheses is reduced considerably. The algorithms are tested and compared in a simulation study. A challenging ground target scenario consisting of road networks and passive sensors is used to evaluate the behaviour of the tracking filters.