Efficient Multitarget Particle Filters for Ground Target Tracking

Many factors make the ground target tracking problem decidedly nonlinear and non-Gaussian. Some of these factors include the relatively poor angular accuracy of GMTI sensors, the presence of persistent clutter and target obscuration, and the complexity of target maneuvers. Because these difficulties can lead to a multimodal posterior density, a Bayesian filtering solution is more appropriate than a point estimate. In the last decade, the particle filter has emerged as a Bayesian inference technique that is both powerful and simple to implement. The price for this flexibility is almost entirely computational; particle filter run-times can be two orders of magnitude longer than those of Kalman filter variants. In this work, we propose to investigate the design of efficient particle filters for multitarget ground tracking. Using simulated data, we will consider multitarget scenarios involving on-road targets, off-road targets, on-road/off-road transitions, and move-stop-move cycles. Instances of uncertain data association will be produced using closely-spaced targets, stopped targets, clutter, and road intersections. An efficient design will be achieved by adaptively managing the type of particle filters, the number of particles, and the enumeration of hypotheses during data association.