A Unified Bayesian Approach to Nonlinearity in Multitarget Tracking

ABSTRACT: Nonlinearity in target tracking poses a far more difficult challenge than is usually understood. Besides nonlinear sensor and target-motion models, one must also address"nonstandard"sources of nonlinearity: known or unknown sensor fields of view; known or unknown clutter; target appearance and disappearance, etc. Furthermore, multitarget tracking (MTT) algorithms are inherently nonlinear. Any systematic MTT analysis must consider all of these forms of nonlinearity. Scientific Systems Company, Inc. proposes a theoretically foundational approach to multitarget nonlinearity analysis. First, we will devise computationally tractable nonlinearity figures of merit for MTT, using theoretically rigorous methods. Second, we will exploit the fact that the multitarget recursive Bayes filter inherently accounts for the above forms of nonlinearity. Specifically, we will develop approximations of the multitarget Bayes filter that comprehensively incorporate the above nonlinear models. We will also investigate new approximate filters that can operate in unknown backgrounds, including unknown clutter and unknown detection profiles. We will implement these algorithms using Gaussian mixture and/or particle-based techniques, and test their sensitivities to the various forms of nonlinearity just mentioned. The project team includes Dr. Ronald Mahler of Lockheed Martin. Lockheed Martin will provide both technical and commercialization support in the application of multitarget tracking technologies. BENEFIT: Unified nonlinearity-resistant multitarget tracking (MTT) algorithms are of major interest to all branches of the military. Commercial application includes law enforcement, industrial and homeland security, air traffic control, and weather radar applications.