ECE Seminar with Dr. Leonid Perlovsky

Starts:
4:00 pm on Wednesday, September 11, 2013
Location:
Photonics Center, 8 Saint Mary’s St., Room 339
URL:
http://www.bu.edu/ece/files/2013/09/Perlovsky.pdf
Maximum Likelihood Joint Tracking and Association in a Strong Clutter (“Track-Before-Detect”)

With Dr. Leonid Perlovsky
Principal Research Scientist
Sensors Directorate of the Air Force Research Laboratory
Wright Patterson Air Force Base, Ohio


Faculty Host: David Castanon

Refreshments will be served outside Room 339 at 3:45 p.m.

Abstract: Performance of the state-of-the-art algorithms for tracking in strong clutter is significantly below the information-theoretic limit as indicated by the Cramer-Rao Bound (CRB) for tracking in clutter. The reason for this underperformance is combinatorial complexity of algorithms. When clutter is strong, so that target signals are below clutter, association of signals with tracks and track estimation have to be performed concurrently. First, a tracking algorithm has to decide which signals come from tracked objects and which come from clutter. And second, track parameters (position, velocity, etc.) have to be estimated. The only information to solve the first problem comes from the second one: track signals form a reasonable track. Therefore, both parts of the problem have to be solved concurrently. This requires considering multiple associations between data and tracks concurrently with detections; in other words, detection, association, and tracking should be performed jointly. The number of associations grows combinatorially with the number of data points (every point could be associated with many points in the next scan and each point could be associated with many points again and again). Therefore, performance is limited by complexity of computations rather than by information in the data. I present a non-combinatorial solution of the maximum likelihood (ML) joint tracking and association problem resulting in a significantly improved performance. Contributions of this work include the ML formulation of joint detection, association, and tracking problem, and its non-combinatorial solution, a general mathematical approach to solving similar problems that used to be considered combinatorially complex. I also discuss intuitively what enabled this improvement. I briefly mention future research directions.

About the Speaker: Dr. Leonid Perlovsky is a Principal Research Scientist with the Sensors Directorate of the Air Force Research Laboratory at Wright Patterson Air Force Base, Ohio. He has a Ph.D. in Theoretical and Mathematical Physics from the Joint Institute for Nuclear Research in Moscow, Russia. His research interests span signal processing, bio-inspired algorithms for detection, tracking, fusion, recognition, situational awareness, data mining, cybersecurity, mathematical models of the mind including language, emotional intelligence, computational intelligence, semantic systems, and semiotics. He is the author of four books on algorithms, intelligent systems, and workings of the mind. He is Editor-in-Chief of the international journal, Physics of Life Reviews. He is the recipient of numerous awards, including the Gabor Award from the International Neural Network Society and the McLucas Award in 2007 from the US Air Force.