Tracking a Coordinated Group using Expectation Maximisation

Roslyn Lau (ANU)

COMPUTER VISION AND ROBOTICS SERIES

DATE: 2013-02-28
TIME: 16:00:00 - 17:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
A common assumption in multiple object tracking is each object moves independently of other objects; however, there are cases where objects travel in a group. In these cases, there are dependencies between the objects; consequently, group object tracking should be used. We investigate efficient approximate methods to produce real-time group object tracking algorithms. Specifically, an algorithm based on expectation maximisation (EM) and belief propagation (BP) is introduced to track a single group of objects where the number of objects is known. The group expectation maximisation belief propagation (GEMBP) filter is compared to the joint probabilistic data association (JPDA) filter, the probabilistic multiple hypothesis tracker (PMHT) and the group PMHT for scenarios involving missed detections and clutter. We introduce a variant of the optimal subpattern assignment (OSPA) metric that includes a penalty for track swaps; results show that the GEMBP filter exhibits fewer track swaps than the JPDA filter.
BIO: