Persistent wide area sensor coverage enables unique intelligence analytic capabilities such as pattern-of-life detection, unsupervised pattern discovery, and anomaly detection. As these capabilities incorporate machine learning and artificial intelligence techniques, large datasets are necessary for training and validation. However, the lack of datasets with high fidelity dynamic targets and actors, along with the high cost of annotating live ground truth experiments impacts the development of these solutions. To address these challenges, we introduce TrailBlazer: a GAN trained high-fidelity track simulator. Trailblazer development has two primary goals: (1) to collect a set of high-quality real activity data for use in validating simulated tracks and for final tracker testing and (2) to develop a high-fidelity track simulator trained on our real validation data. TrailBlazers high fidelity real and simulated data will enable the application of modern, data intensive machine learning techniques, such as deep learning, to intelligence, surveillance and reconnaissance (ISR) tracking and analytics.
