Department of Electrical and Computer Engineering ¹
Villanova University
Villanova, PA 19085

Trellis Structure Approach to Multitarget Tracking

Richard Perry - Anand Vaddiraju - Kevin Buckley

perry(anand, buckley)@ece.vill.edu

Abstract:

In this paper we address sequential multitarget tracking for Radar applications. Specifically, we consider the problem of associating location measurement data over time to form estimates of multiple tracks. The algorithmic approach we employ is sequential and very flexible in that it can handle missed detections, false alarms, track initiation, and number-of-track estimation, while employing maximum likelihood or Bayesian cost functions. The approach is based on a trellis diagram which depicts the possible progressions of sequences of location measurements over time. Computation requirements are managed using a generalized K-path, L-list Viterbi algorithm to prune candidate track sets which have prohibitively high costs. Here we focus on a Bayesian solution to the joint problem of estimating both the number-of-tracks and the tracks themselves.