Aptima’s Software Helps Military Detect Threats
Aptima Develops ‘VISION’ for DARPA – Software to Help Military Detect Threats in Video
Decision support tool recognizes behaviors and adversarial intent from wide-area motion imagery
Woburn, MA, 24 October 2011 – A picture may be worth a thousand words, yet it’s the millions of images pieced together from surveillance video that provides U.S. troops the most valuable intelligence. Airborne video can provide early-warning indicators and improved situational awareness, yet without new technologies, human analysts have reached limits in the ability to uncover threats quickly enough to be acted upon.
To overcome this limitation, Aptima, which applies expertise in human-centered engineering, is developing VISION, ‘Video-based Identification of Structured Intent for Objects and Networks,’ for the Defense Advanced Research Projects Agency (DARPA). As an intelligent decision support tool, VISION will automatically scour huge volumes of wide-area video taken from UAVs and other airborne platforms.
Using statistical algorithms, VISION can learn and detect the activities, interactions, and functions of static and moving objects within motion imagery. This search and analysis technology is expectedto play a key role in real-time threat detectionby providing commanders and their forces with knowledge about the activities and locations of adversaries.
How VISION works
While individual actions in a city or village may not convey much meaning or raise suspicions separately, the collective behaviors and interactions of people and vehicles leave semantic traces (‘footprints’) related to their roles and activities, and the functions of the places they visit.
An innocuous-appearing shop for example, may have multiple people going to and from at odd hours, with trucks making U-turn maneuvers and intermittent stops. Although benignly suggesting pizza delivery, those complex patterns of activities and tracks over time and geographic area could indicate to analysts that the building and its off-hour visitors are assembling and delivering IEDs, not pizzas.
The advantage of wide-area motion imagery (WAMI) lies in the ability to monitor huge swaths of geography up to 40 square miles for hours to days at a time, making it difficult for adversaries to disguise their activities. This also creates an extreme workload for human analysts relying on current manual techniques. By taking hours and days to wade through, assess, and exploit terabytes of surveillance video, critical battlefield questions go unanswered and timely threat cues are missed. VISION’s algorithms can learn the signatures of normal and abnormal behaviors of people and vehicles, applying these patterns to identify and classify these entities over time.
“Rather than poring through millions of video images, we expect analysts to use VISION in the same way that large repositories of text can be searched and indexed,” said Georgiy Levchuk, a Simulation & Optimization Engineer and Aptima’s VISION program manager. ”The algorithms will allow analysts to make simple and complex queries to watch for or search the imagery for particular events, behaviors, and relationships. VISION can also be used forensically to discover activities of interest of people and vehicles within huge volumes of previously collected data.”
Through a previous contract with DARPA, Aptima completed research that validates the ability of algorithms to identify various types of behaviors and activities. Read the entire article Algorithms for behavior recognition in wide-area video of urban environments at SPIE.org.
DISTRIBUTION STATEMENT A. Approved for public release; distribution is unlimited. (DISTAR Case 16665 Approved)