Publisher's Synopsis
A methodology for the context-aided tracking of ground vehicles in remote airborne imagery is developed in which a background model is inferred from hyperspectral imagery. The materials comprising the background of a scene are remotely identified and lead to this model. Two model formation processes are developed: a manual method, and method that exploits an emerging adaptive, multiple-object-spectrometer instrument. A semi-automated background modeling approach is shown to arrive at a reasonable background model with minimal operator intervention. A novel, adaptive, and autonomous approach uses a new type of adaptive hyperspectral sensor, and converges to a 66% correct background model in 5% the time of the baseline { a 95% reduction in sensor acquisition time. A multiple-hypothesis-tracker is incorporated, which utilizes background statistics to form track costs and associated track maintenance thresholds. The context-aided system is demonstrated in a high-fidelity tracking testbed, and reduces track identity error by 30%.
