Exploitation of Hyperspectral Infrared Imagery for Persistent Tracking

New tracking algorithms and sensing modalities are required to enable long duration (>1 hour) tracking of vehicles in urban environments. Anecdotal evidence suggests that long wavelength infrared (LWIR, lambda~8 to 12 microm) spectra of vehicle paints are sufficiently distinct to facilitate spectral-matching-based object detection and thereby reduce object confusion by tracking algorithms. Physical Sciences Inc. (PSI) has also observed that LWIR spectra of vehicle paints are sensitive to contamination from oil, dust, and other foreign materials. This suggests that nominally identical vehicles, i.e., same make, model, and paint color, may be spectrally distinct as a result of their environmental history. In the Phase I program PSI will conduct field experiments and implement detection and tracking algorithms to quantify the value of LWIR hyperspectral imagery relative to broadband LWIR imagery for unsupervised object detection and long duration tracking. Quantitative assessment of LWIR-based object detection will be made using Receiver Operator Characteristic (ROC) curves. PSI will also demonstrate spatial resolution enhancement of hyperspectral LWIR imagery via fusion with resolution panchromatic imagery and assess the impact of resolution enhancement on ROC curves.