Target tracking by symbiotic particle filtering

In the past decade and a half, particle filtering (PF), has gained considerable popularity in dealing with nonlinear and/or non-Gaussian target tracking problems. However, in problems of high dimensionality, i.e., when many targets are present in the field, a very large number of particles is required for satisfactory performance of the methodology. In this paper we improve our previously proposed multiple particle filter scheme by introducing "symbiosis" among the particles filters. In other words, we allow the individual particle filters, when necessary, to combine their random measures and form a new random measure with particles of high dimensions, or a single particle filter to split into one or more filters with particles of smaller dimensions. We validate the method on the problem of target tracking in a network of acoustic sensors.