Computers may be ferocious number-crunchers, but they can't match humans for eyesight. The simple task of spotting an object in a scene can take hours on a powerful desktop machine. On Dec. 1, researchers at the David Sarnoff Research Center in Princeton, N.J., unveiled a system that dramatically boosts the ability of computers to process images. "This really marks the beginning of a new era of real-time computer vision," boasts Sarnoff President James E. Carnes.
The system--a combination of hardware and software--begins the processing with a simple, low-resolution image. That way, it can home in on key features--such as a moving tank or a person--without having to search through millions of pixels. "The key is to ignore extraneous detail," explains Sarnoff Vice-President Curtis R. Carlson. Sarnoff researchers envision such applications as reading X-rays, providing sight for factory-floor robots, and enabling PCs to handle real-time video processing. Sarnoff has formed a new company, Sensar, to bring the technology to market.