Threat Detection Using Artificial Intelligence and Machine Learning

Friendly forces at forward operating bases and expeditionary airfields require autonomous solutions to alerts of potential threats (dismounts, drones, etc.). For such a system to be useful, it needs to operate at high probability of detection and low probability of false alarm. Several technical challenges exist, including the variability of environments and threat types, the presence of background activity, and the need for an easy set-up process. To address these challenges, we propose to develop a Layered Omnidirectional Threat Tracking and Alert system (LOTTA). LOTTA uses a scalable, heterogeneous suite of complimentary sensors (EO/IR, GMTI and UGS) to detect threats of varying type (humans, vehicles, quad-copters, etc.) and behaviors (moving and stationary). Next, LOTTA uses STRs Multi-Stage Multiple-Hypothesis Tracker (MS-MHT) to fuse detections from complementary sensors and track these potential threats. LOTTA uses state-of-the art machine learning techniques to classify target types, and detect attributes and behaviors that could indicate a threat (e.g., a human carrying a weapon, or a vehicle driving a certain way toward the asset). Finally, LOTTA provides a user interface for easy sensor configuration, threat definition (e.g., drawing of a tripwire) and threat alert.