Distributed data association for Multi-target tracking in sensor networks

Abstract

In this paper, we explore the problem of tracking multiple targets through a field of sensors. Each sensor node is capable of making noisy measurements of the targets¿ positions, performing on-board computation, and wirelessly transmitting information to neighboring nodes. The problem of multitarget tracking (MTT) can be decomposed into two main fusion problems: estimation and data association. Using Kalman-consensus filtering (KCF), introduced by Olfati-Saber, the authors have recently addressed distributed estimation in tracking for a single target. Data association techniques for multitarget tracking are categorized by how many time-indexed sets of measurements are made before the associations are considered ¿fixed¿. Most multitarget tracking algorithms perform the data association at a central processing node through a multiple-scan method such as multiple hypothesis tracking (MHT), or single-scan techniques such as joint probabilistic data association (JPDA), Markov Chain Monte Carlo methods (MCMC), or optimal graph matching. Here, the main contribution is to introduce data association algorithms for ¿distributed¿ multitarget tracking. A formulation of joint probabilistic data association for Kalman-Consensus Filtering is formally derived. Simulations are provided to demonstrate the effectiveness of our distributed multitarget tracking algorithm for tracking multiple maneuvering targets in the sensing environment of a sensor network with 25 nodes.