Minhtri Ho

Performance Improvement of SME Filter for Multiple Target Tracking

Tuesday, May 25, 1999
2:00 PM
406 Egan

Abstract

Tracking multiple targets is a difficult problem, requiring first the correct association of data with targets, and then the estimation of tracks from the noisy data. The data association step is often difficult, especially when targets are maneuvering in close proximity and have frequently crossing trajectories. In the Symmetrical Measurement Equation (SME) approach to tracking, the data association and filtering steps are combined and Extended Kalman Filter (EKF) is used. This approach is conceptually and computationally simple, but unfortunately suffers from instability with large initialization error and at trajectory crossings, especially when targets are maneuvering. Therefore, there is a need for a method of initializing SME that prevents the filter from entering unstable states, and also a need for an improved algorithm for SME filter that is more robust at track crossings. In this thesis, we focus our effort on the initialization problem. We thoroughly study how initialization affects the performance of Kalman filter in general and SME filter in particular. Existing methods of initializing Kalman filter are studied and their use in derivation of an initialization method for SME filter is examined. Consequently, we propose a practical initialization method that makes use of the information obtained from the first set of measurements to initialize the SME filter. The practical initialization method is tested with computer simulations for a single run as well as Monte Carlo runs, and is observed to be beneficial for the performance of SME filter. A preliminary work on Proportional Integral (PI) EKF for SME algorithm is also carried out in our work, and results show that the performance of Proportional Integral-EKF for SME algorithm is beneficial to the stability of the SME tracker at track crossings.

Thesis Committee:
Prof. Bahram Shafai (advisor)
Prof. Hanoch Lev-Ari
Prof. Michael Malioutov