Tracking multiple targets with a sensor network

Abstract

The problem of tracking multiple targets moving through a network of sensors is considered. It is assumed that the sensors send regular returns to a central node at which all processing is performed. Two approaches to the problem are considered: the unscented Kalman filter and a simple implementation of the auxiliary particle filter. The algorithms are formulated under a general sensor model which does not assume a particular statistical model for the measurements. Monte Carlo simulations are used to assess the performances of the algorithms with both a binary sensor model and a non-thresholded sensor model. The unscented Kalman filter significantly outperforms the particle filter in both cases and has a much lower computational expense