Mutiswarm particle filter for robust tracking under observation ambiguity

Particle Filters are a traditional optimization tool for nonlinear, non-Gaussian dynamic-state estimation such as visual tracking. The particle filters, however, suffer from particle degeneracy problem which is caused by the mismatch between the proposal distribution and the target distribution. In this paper, we propose a method for improving the performance of the particle filter via multiswarm-based Particle Swarm Optimization (PSO). We utilize PSO to obtain samples that are well matched with the likelihood distribution, and its converging property is handled with the exclusion between particles. Additionally, we incorporate multiswarm algorithm in the PSO combined particle filter to deal with ambiguities in estimation task. The resulting filter is applied to the object tracking problem with ambiguous observations, and its performance is tested. We present the experimental results that demonstrate improved accuracy with the same or less computational cost.