Innovative Filtering Techniques for Ground Target Tracking

The project focus is the development of a ground target tracking algorithm to support the JSTARS radar operating in the GMTI mode. Multiple hypothesis tracking (MHT) algorithms maintain alternative data associations to represent report-to-track association ambiguities. In Phase 1 we developed a prototype MHT algorithm not requiring the Bayes’ posterior distribution for the target state to be Gaussian. State estimation is accomplished using the deformation method, a form of particle filtering (sequential Monte Carlo state estimation) that requires neither hypothesis pruning nor hypothesis regeneration. The deformation method transforms existing target state estimates into samples that closely approximate draws from the Bayes’ posterior distribution for the target state. Data association is treated as an assignment problem and is solved using the Munkres algorithm. The smooth deformation of the continuous target state variables is through the action of a stochastic differential equation. The discontinuous deformation of the discrete target state variables is through the application of the Metropolis algorithm. In Phase 2 we will incorporate the processing of target signature information from high range resolution (HRR) data. We also will model the effects of terrain on vehicle motion and observability. Certain computations will be parallelized to reduce the algorithm execution time.