A consensus/fusion based distributed implementation of the particle filter

In view of non-linear, non-Gaussian tracking applications in sensor networks, we propose a consensus/fusion based, distributed implementation of the particle filter (CF/DPF). The proposed distributed implementation addresses three important issues: (i) Extending the idea of the channel filters , separate fusion filters are designed to consistently assimilate the local filtering distributions into global posterior by compensating for the common past information between neighbouring nodes typically overlooked in consensus-based distributed particle filter implementations. (ii) The proposed method is not limited to Gaussian approximation for the global posterior density. (iii) Finally, the condition that the consensus step converges between two consecutive observations is partially relaxed. Our numerical simulations verify that the output of the CF/DPF is almost identical to that of the centralized particle filter.