Bayesian cell filter for constrained non-Gaussian estimation

The Bayesian approach provides the most general formulation of the recursive state estimation problem. Except for linear-Gaussian systems, the solution is seldom amenable to implementation. This paper poses the estimation problem in discretized state space. A novel approach is used to model probabilistic dynamics as finite state Markov chains. The Bayesian cell filter can handle nonlinearities, nonGaussian process and measurement noise and constraints. The filter splits the problem into offline modeling and online estimation tasks. The cell filter is compared with Monte Carlo based particle filter for accuracy and efficiency.