Can a computer spot a criminal or a terrorist infiltrating an airport tarmac? Face-recognition programs are useless if a person's face is obscured, unlit, or just too far away. So a number of researchers are trying to identify a person simply by the way he or she walks.
One approach is to collect video images of the surveillance area and analyze everybody passing through. "We've been trying to measure things like stride length and body shapes," says Aaron Bobick, a computer vision researcher at Georgia Tech in Atlanta. Video cameras still require good light and a clear view, however, so Gene Greneker at the Georgia Tech Research Institute is experimenting with radar. "It can see through clothes, at night, at long ranges," he says.
Greneker and his team use a device that sends out a signal and measures the echo. If the return waves shift to higher frequencies, that means they reflected off something approaching. "There are different shifts for different body parts, because they are moving at different velocities," Greneker says. A computer program analyzes these shifts and creates a radar fingerprint for each person's walk. So far researchers have collected gait profiles of about 100 test subjects, which the program can identify up to 80 percent of the time. That is not nearly good enough for airport security, but Bobick is convinced gait recognition can be made to work: "If I show you moving light displays of people walking, you can distinguish one person from the next. We know the information is there."
