Joint Target Tracking and Classification using GMTI/HRR Data

High-Resolution Radar (HRR) range profiles provide a means to improve a GMTI tracker's performance when vehicles enter kinematically ambiguous situations. There is a need for methods that process HRR profiles on-the-fly, enhance their features, discover the salient features, and learn robust representations from a small number of views. In Phase I we demonstrated feasibility of pattern enhancement and learning methods based on neural models of human visual processing, learning, and recognition that provide a novel approach to address this need. Promising simulation results using datasets of military targets indicate that these methods can be exploited in the field to improve the association between GMTI detections and tracks. The proposed effort will extend and integrate the system designed in Phase I to create a prototype feature-aided tracking system, providing coupled solutions to the problems of GMTI tracking and on-the-fly HRR learning & recognition. We will analyze and enhance stability of HRR feature discovery and learning & recognition, and enable the incorporation of a priori knowledge. We will demonstrate and transfer to AFRL/SN the integrated feature aided tracking system and results on scenarios of interest. The proposed strategy has potential for real-time processing, which will be implemented on COTS signal processing boards.