Hyperspectral Identification for Collaborative Tracking

Many problems exist, particularly in defense and security scenarios, in which long-term tracking of objects is important. Long-term tracking generally requires the use of collected features in order to uniquely identify the object of interest. As conditions under which the features are collected changes, so do the features. While different sensor phenomenologies have been developed to collect feature measurements (i.e., measurements that depend on target attributes such as shape or color), this effort is focused on using dynamically collected features from hyperspectral sensors under changing operating conditions to extend track lifetime. Changing operating conditions preclude the use of a priori feature databases; thus, a feature database that is built “on-the-fly” is required to reliably track vehicles over longer periods. Furthermore, the feature database must be dynamically managed since, as conditions that affect the collected features change, the previously collected features may no longer be used to realiably identify the vehicle. Toyon Research proposes to analyze and statistically model hyperspectral feature data, and develop a feature database management algorithm to accommodate changing operating conditions. Moreover, Toyon will analyze the utility of using hyperspectral signatures and the database management algorithm to extend track life.