We will demonstrate the feasibility of detecting tactically meaningful complex events in sensor input streams using an efficient pattern matching technology embodied in the Soar cognitive architecture. Our focus in Phase I will be on video streams, such as those that might be produced by unattended ground sensors or unmanned aerial systems. To reduce risk, we propose devoting a portion of our effort to the problem of reliably producing simple or atomic event markers from the video stream, so that our complex event detection capability has a solid foundation of input data.
Benefit: Navy and Marine Corps personnel need to anticipate terrorist actions. Threatening complex behavior is a hostile act that requires a series of human actions be taken in advance. The approach proposed here could reduce personnel requirements by providing producing actionable intelligence through automated monitoring. The approach also has significant commercial applicability outside the defense environment, with applications in public and private security, including specialized fields such as casinos. Other small but demanding markets include child care and animal and human behavior testing.
Keywords: video extraction, video extraction, complex event, Pattern Matching, event detection
