Multi-Target Tracking Using the Random Finite Set Theory and Bayesian Framework

Seyed Hamid Rezatofighi (ANU)

APPLIED SIGNAL PROCESSING SERIES

DATE: 2013-11-14
TIME: 10:00:00 - 11:00:00
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:
Despite significant technical advances made in automatically tracking multiple targets, this problem remains a challenging task in many practical applications due to their complex nature. Tracking several pedestrians in a very crowded scene in surveillance camera, populated and dense cells in biological sequences and multiple similar targets in very noisy sequences of sonar and radar can be counted as some examples of these applications. The main challenge in these applications is to estimate of the time-varying number and the state of targets based on a set of uncertain measurements. Often, not all of the targets are detected by the sensor due to target occlusion or the performance of sensor. Moreover, the observations generally include a set of spurious measurements (clutter) not originating from any target. The behaviors of targets such as maneuvering motions, entering, exiting or temporarily disappearing from scene and interactions with other targets (i.e., targets splitting and merging) add more complexities. An automated tracking system should be able to track an unknown and time-varying number of targets in the presence of data association uncertainty, clutter noise, and detection uncertainty.

Bayesian tracking approaches are a class of probabilistic tracking algorithms that have become popular for many tracking applications in recent years. However, the traditional Bayesian tracking algorithms such as Kalman and Particle filters are basically designed for single target or constant cardinality multi-targets. In order to apply these approaches for tracking time-varying number of targets, they require additional data association and track management techniques. As a result, the complexity and the performance of tracking algorithms depend not only on the tracking filter but also the data association and track management algorithms. As a unified alternative, a new generation of Bayesian filters based on Random Finite Set (RFS) theory, e.g. the Probability Hypothesis Density (PHD), Cardinalized PHD (CPHD) and Multi-Target Multi-Bernoulli (MeMBer) filters has been proposed in the literature. In this talk, a brief introduction to these random finite set based Bayesian filters will be presented.