Particle filtering with adaptive number of particles

Nonlinear/non-Gaussian dynamic systems can be tackled by a number of filtering methods. We are interested in particle filters, which perform a discrete characterization of the posterior distribution of the system based on a random set of points. The dimension of the random set is a design issue and typically large values are required to ensure proper tracking of the system. This is typically solved by a worst-case criterion, involving a waste of computational resources. In this paper we are interested in the design of a particle filtering algorithm which is able to adapt the dimension of its particle pool. The new filter, which uses information from the innovation error to modify the number of particle to use, has shown remarkable results in terms of performance and computational cost reduction.