Innovative Filtering Techniques for Ground Target Tracking

In multi-target tracking applications, ambiguities generally arise in how to associate sensor reports with targets. Multiple hypothesis tracking (MHT) algorithms maintain multiple alternative data associations to represent these ambiguities. We propose to develop an (MHT) algorithm design that will provide improved ambiguity resolution and target track estimation accuracy, and therefore improve long-term track maintenance capability. The methods we propose apply generally to the tracking of ground targets and to sensing technologies across the EM spectrum. The basic statistical estimation framework we propose is that of particle filtering (sequential Monte Carlo state estimation). The extension of existing data association hypotheses to incorporate additional scans of sensor data is cast as an assignment problem and is solved using the Munkres algorithm. Hypothesis pruning and hypothesis merging are implicit and are eliminated as formal hypothesis management operations. The updating of estimated target tracks is accomplished through the application of a stochastic differential equation to the continuous target state variables and the application of the Metropolis algorithm to the discrete target state variables. The effect is that through a deformation process existing target state estimates are transformed into samples from the modified Bayes' posterior distribution. The deformation function also provides a mechanism for restructuring hypotheses and target tracks in response to information indicating that modification of past data associations is required. The required computations are well suited for parallel processing at multiple levels. A Phase I task will be to identify the specific elements of the computation that are parallelizable and to quantify the reduction in execution time that can be achieved through parallel computation.