Renyi Entropy based Target Tracking in Mobile Sensor Networks

This paper proposes an entropy based target tracking approach for mobile sensor networks. The proposed tracking algorithm runs a target state estimation stage and a motion control stage alternatively. A distributed particle filter is developed to estimate the target position in the first stage. This distributed particle filter does not require to transmit the weighted particles from one sensor node to another. Instead, a Gaussian mixture model is formulated to approximate the posterior distribution represented by the weighted particles via an EM algorithm. The EM algorithm is developed in a distributed form to compute the parameters of Gaussian mixture model via local communication, which leads to the distributed implementation of the particle filter. A flocking controller is developed to control the mobile sensor nodes to track the target in the second stage. The flocking control algorithm includes three components. Collision avoidance component is based on the design of a separation potential function. Alignment component is based on a consensus algorithm. Navigation component is based on the minimization of an quadratic Renyi entropy. The quadratic Renyi entropy of Gaussian mixture model has an analytical expression so that its optimization is feasible in mobile sensor networks. The proposed active tracking algorithm is tested in simulation.