Behavioral Discrimination of Moving Targets in Ground Moving Target Indicator (GMTI) Radar

Black River Systems, with notable GMTI radar/tracking experience coupled with our recent work to characterize and model dismounts, proposes an innovative algorithmic approach to discriminate between humans and wildlife based on motion analysis. We developed the BRAT Tracker used on a daily basis to process in-theater GMTI operational data to generate GMTI tracks. We will process STANAG 4607 GMTI dots through the BRAT to develop GMTI Tracks. Using a Hartigan Cluster-Leader algorithm, Track Clustering forms groups of nearby GMTI Tracks. For each group, Feature Extraction computes features, including size, motion, Radar Cross Section, and associated statics (mean, covariance, minimum, maximum), that characterize group motion. Principal Component Analysis identifies features that are correlated, contain the most information, and determines a mapping from the original feature space to a reduced dimensionality space. These reduced features are processed through Track Discriminator by a Support Vector Machine to provide (nonlinear) classification of the group as Dismount, Animal, Clutter, or Vehicle. We also investigate a Relevance Vector Machine as the Track Discriminator, and compare its performance with the SVM. We will conduct parametric studies (track length, revisit rate, features, and resolution) to provide guidance for operating conditions (sensor accuracies and required 4607 fields).