Drift homotopy particle filter for non-Gaussian multi-target tracking

In this paper, we present a novel particle filtering algorithm for multi-target tracking problem in a non-Gaussian environment. Our approach incorporates a Markov Chain Monte Carlo scheme with drift homotopy after an appropriately modified resampling step. The algorithm is tested on a multi-target tracking model with a linear and a nonlinear observation model. Both targets dynamics model and observation model are perturbed by non-Gaussian noises. The results of the numerical tests based on synthetic data indicate that our method significantly improves the performance of the generic particle filter.