State-Space Partitioning Schemes in Multiple Particle Filtering for Improved Accuracy

Multiple particle filtering was proposed as an alternative to particle filtering when tracking states in systems of high dimensions. Multiple particle filters are comprised of a network of particle filters assigned to track subsets of the state, and require each other's obtained information to carry out the filtering. Many improvements of multiple particle filtering have been proposed, however, there have been only a few efforts studying the effects of state partitioning on the filtering performance. In this paper, we propose two novel partitioning schemes for improved accuracy of multiple particle filtering based on: i) random permutations, and ii) the connectedness of the states, i.e. the topology of the system. Computer simulations show that the filter significantly benefits from state permutations, especially when driven by the information in the topology.