Distributed particle filtering in the presence of mutually correlated sensor noises

We propose two distributed particle filter (DPF) algorithms for sensor networks with mutually correlated measurement noises at different sensors. With both algorithms, each sensor runs a local particle filter that knows the global (all-sensors) likelihood function and is thus able to compute a global state estimate based on the measurements of all sensors. We propose two alternative distributed, consensus-based methods for computing the global likelihood function at each sensor. Simulation results for a target tracking problem demonstrate that both DPF algorithms exhibit excellent performance, however with very different communications requirements.