Abstract
 Multiple target tracking answers questions such as “How many targets are
there?”, “Where are they now?”, and “Where are they going?”. This requires the
processing of large amounts of data arriving from one or more sensors in a
computationally efficient manner.
One of the common assumptions of multiple target tracking is that targets
travel independently; however, there are several cases where targets travel in a
group. For example, a convoy of vehicles may travel on a road, a fleet of ships
may sail across the ocean, and a group of planes may travel in formation.
Tracking a coordinated group of targets using radar detections is challenging
because filters can incorrectly produce tracks that switch identities (track
swaps), or tracks can incorrectly converge to the same estimate (track
coalescence).
In this paper, we present a unified model to track a single group of targets in
which the number of targets is varying and unknown. Specifically, we extend the
virtual leader model to formulate the group of targets and apply structured mean
field approximations to produce a computationally efficient algorithm. By using
the dependent motion model and structured mean field approach, we are able to
reduce track swaps and coalescence.
Results for simulated data show the algorithm to be accurate for various values
of probability of detection, false alarm rate, and measurement noise. Future
extensions include tracking multiple groups of targets and further reducing the
computational complexity.
