The Economist
October 26, 2008
Monitoring surveillance cameras is tedious work. Even if you are concentrating, identifying suspicious behaviour is hard. Suppose a nondescript man descends to a subway platform several times over the course of a few days without getting on a train. Is that suspicious? Possibly. Is the average security guard going to notice? Probably not. A good example, then—if a fictional one—of why many people would like to develop intelligent computerised surveillance systems.
- A d v e r t i s e m e n t
The perceived need for such systems is stimulating the development of devices that can both recognise people and objects and also detect suspicious behaviour. Much of this technology remains, for the moment, in laboratories. But Charles Cohen, the boss of Cybernet Systems, a firm based in Ann Arbor, Michigan, which is working for America’s Army Research Laboratory, says behaviour-recognition systems are getting good, and are already deployed at some security checkpoints.
Human gaits, for example, can provide a lot of information about people’s intentions. At the American Army’s Aberdeen Proving Ground in Maryland, a team of gait analysts and psychologists led by Frank Morelli study video, much of it conveniently posted on the internet by insurgents in Afghanistan and Iraq. They use special object-recognition software to lock onto particular features of a video recording (a person’s knees or elbow joints, for example) and follow them around. Correlating those movements with consequences, such as the throwing of a bomb, allows them to develop computer models that link posture and consequence reasonably reliably. The system can, for example, pick out a person in a crowd who is carrying a concealed package with the weight of a large explosives belt. According to Mr Morelli, the army plans to deploy the system at military checkpoints, on vehicles and at embassy perimeters.
The Emergency Election Sale is now live! Get 30% to 60% off our most popular products today!