Sep. 19, 2012 at 9:00 PM ET
The military has been looking into securing borders with automated surveillance systems, but the cameras and computers have trouble distinguishing between terrorists and coyotes — so a new project uses the most powerful visual threat analysis engine available: the human brain.
Computer vision is growing more powerful and useful every year: it powers autonomous cars, facial recognition in cameras, and many other practical applications. But a false positive on a camera's smile shutter isn't a big deal, while a false positive along a contested border could cause great expense or even armed conflict.
The Cognitive Technology Threat Warning System (or CT2WS, hardly less cumbersome) used a visual processing system to watch terrain in tests, and it returned a staggering 810 false alarms every hour, almost a third of the total events it recorded. Spread those numbers over miles and miles of forested or tropical borders, and the problem soon gets out of control.
In order to reduce the number of false alarms, the researchers (spread over a number of institutions but led by DARPA's Gill Pratt) inserted a human into the equation. But not just someone looking at the camera feed for hours; such monotonous work is fatiguing and often leads to missed events.
Instead, the screen shows image after image in quick succession to the user, while monitoring their brainwaves for a specific pattern: called a P-300 brainwave, it indicates something that needs to be paid attention to by the visual system. And humans, trained for years in the real world, are incredibly responsive to even slight glimpses of motion and suspicious shapes.
The result was a more than 99 percent reduction of false alarms: the researchers' testing showed just five per hour, and when combined with an extra radar system, that number dropped to zero.
It certainly may sound a bit alarming, using a person as essentially a graphics processing unit, but the fact is that people are naturally good at this and even if CT2WS is never deployed, computer vision systems can learn a lot from the way humans do their job.
via Ars Technica
Devin Coldewey is a contributing writer for NBC News Digital. His personal website is coldewey.cc.