Learn-while-tracking, feature discovery and fusion of high-resolution radar range profiles
Richard Ivey, Allen M. Waxman, David A. Fay, David P. Martin · 2003
High-Resolution Radar (HRR) range profile data, obtained simultaneously with radar GMTI detection of a moving target, can provide a means to improve a target tracker 's perj4ormance when multiple vehicles are in kinematically ambiguous situations. There is a need for methods that can process HRR profiles on-the-fly, enhance their features, discover which of the features are salient, and learn robust representations fi-om a small number of views. We present signal processing and pattern recognition methods based on neural models of human visual processing, leaming, and recognition that provide a novel approach to address this need. Promising simulation results using a set of militaly targetsfi-om the MSTAR dataset indicate that these methods can be exploited in the field, on-line, to improve the association between Gh4TI detections and tracks. The approach developed here is extensible and can easily accommodate multiple sensor plaljorms and incorporate other target signatures that may complement the HRR profile, for example spectral imagery of the target. Thus, our approach opens the way to sensor fused target tracking.