Bayesian Troop-Target Detection, Tracking, and Prediction Using FOPEN-GMTI Radar

This proposal addresses the problem of automatically detecting, tracking, and predicting the locations and velocities of formations of dismounted ground troops concealed in foliage, using data collected from P-band FOPEN-GMTI radars. Scientific Systems Company, Inc. (SSCI) and its subcontractor Lockheed Martin Tactical Systems (LMTS) will address this ``troop-target'' problem with a theoretically rigorous generalization of the concept of a ``probability hypothesis surface (PHS),'' first introduced in the mid-1990s as a potential approach for group-target processing under the DARPA DMIF program. Theoretically speaking, we apply a mulitarget nonlinear filter that is a multitarget statistical analog of a constant-gain Kalman filter. It propagates a first-order multitarget moment of the multitarget posterior distribution rather than the full multitarget distribution itself. Rather than attempting to assemble information about a troop formation by separately detecting and tracking its constituent individuals, this filter first detects and tracks only the over-all bulk behavior of a formation that is obscured by clutter and missed detections. What is being tracked at any time is a surface that is an estimate of the geographical shape and density of the troop formation. Only if the quantity and quality of data permits, does the filter then attempt to extract and track the individual constituents of the formation, which appear as distinct peaks in the surface. Our approach should be capable of detecting, extracting, tracking, and predicting the future states of troop formations despite relatively small probabilities of detection and relatively large clutter densities. The project team includes Dr. Ronald Mahler of Lockheed Martin. Lockheed Martin will provide both technical and commercialization support in the application of detection/tracking technologies.