Implementation of the homotopy particle filter in the JPDA and MAP-PF multi-target tracking algorithms

In a conventional particle filter, the information update step can suffer from particle degeneracy if the likelihood function is concentrated on only a few particles. The homotopy particle flow method has been developed to implement the information update in an entirely different manner by using a particle flow function to migrate the particles to regions of the target state space that provide a good representation of the posterior distribution. In this paper we demonstrate how the homotopy particle filter can be implemented with the JPDA and MAP-PF multi-target tracking algorithms, and compare performance to the conventional resampling method.