Maximum Mutual Information Principle for Dynamic Sensor Query Problems

In this paper we study a dynamic sensor selection method for Bayesian filtering problems. In particular we consider the distributed Bayesian Filtering strategy given in [1] and show that the principle of mutual information maximization follows naturally from the expected uncertainty minimization criterion in a Bayesian filtering framework. This equivalence results in a computationally feasible approach to state estimation in sensor networks. We illustrate the application of the proposed dynamic sensor selection method to both discrete and linear Gaussian models for distributed tracking as well as to stationary target localization using acoustic arrays.