Neural net tracks skin color

Our seemingly easy ability to spot and distinguish one object from another is actually a complicated process that evolved over millions of years.

Researchers working to give computers and robots the ability to recognize gestures are up against several challenges. This system must recognize a face or a gesturing hand, and it must be able to continue to distinguish faces and hands as they move around among other objects.

Researchers from the Chinese Academy of Sciences and the Chinese Institute of Modern Optics have come up with a way to use skin color to detect faces and hands. This is trickier than it sounds, because colors are elusive, changing depending on illumination.

The researchers' system uses a camera connected to a processor that uses an artificial neural network to detect skin color, then processes the information further to determine which skin-color objects should be connected together.

The system could be used to enable gaze and gesture control of electronics like appliances and robots. Tests of the system show that it was able to segment gestures with 96.25 percent accuracy, according to the researchers.

The researchers' gesture recognition method could be used practically in five years, according to the researchers. The work appeared in the August, 2003 issue of Optical Engineering.