Date: November 12, 2004
Time: 3:00 pm
Location: GCATT Room 325
Speaker(s): Will Leven
Title: Multiple Target Tracking with Symmetric Measurement Equations Revisited: Unscented Kalman Filters, Particle Filters, and Taylor Series Expansions
Abstract:
The symmetric measurement equation approach to multiple target tracking is revisited using unscented Kalman and particle filters. The characteristics and performance of these filters are compared to the original symmetric measurement equation implementation relying upon an extended Kalman filter. The unscented Kalman filter (UKF) promises more accurate approximation of nonlinearities and simpler implementation of the SME approach than the EKF. The particle filter implementation offers the ability to explore the limits of the SME approach.
The performance of these filters paired with two sets of symmetric measurement equations is analyzed using three criteria: track maintenance, set estimation error, and average mean squared error. Several counterintuitive aspects of the results are discussed, including a previously unknown limitation of the unscented Kalman filter. The point is made that the performance of the SME approach is dependent on the interaction of the set of SME equations and the filter used.
Taylor series expansions are used to evaluate and explain the performance differences between Kalman filter-SME pairings. Using the Taylor series representation, we show how the choice of SME formulation affects the representation, and consequently approximation, of uncertainty in the Kalman filters. This analysis suggests that the Taylor series expansion might be used as a design tool for constructing SME formulations tailored to the tracking situation.

Bio:
Will Leven is currently a Ph.D. student working for Dr. Lanterman at the Georgia Institute of Technology. He earned his M.S. from Georgia Tech in 2004 and graduated from U.C. Berkeley in 2002 with a B.S. in Electrical Engineering and Computer Science.