SSP'05 -- Proceedings

	IEEE/SP 13th workshop on Statistical Signal Processing July, 17-20, 2005 - Bordeaux - France

Information regarding the paper

Title

Bayesian Model Selection for Multisensor Track-to-Track Association and Track Fusion

Author(s)

Huimin Chen	University of New Orleans
X. Rong Li	University of New Orleans

Get the paper in PDF format

To obtain Acrobat Reader (version 5 minimum required) necessary to his read.

Abstract

The problem of track-to-track association and track fusion is studied where the number of targets is unknown. The association of local tracks corresponding to the same target is usually based on a hypothesis testing that yields the desired power. However, there is in general no control of misassociation probability. Here we treat track association as a model selection problem and derives a Bayesian approach that can handle the issue of unknown number of targets which arises in large scale tracking problem where many sensors track many targets. The Bayesian solution provides a posteriori probability for each hypothesis which can be used for track fusion. Another important consequence is that the optimal track fusion algorithm is not linear if noninformative prior for the target state is assumed even when the local track estimates are based on a system with linear Gaussian dynamics. We present a regularized track fusion algorithm which yields smaller estimation error than the existing optimal linear track fusion algorithm. We compare the track association and fusion result with a centralized tracker using a three-sensor two-target tracking scenario.