The characteristics of your walk may not be as distinctive as the swaggering of John Wayne or the sashay of Joan Collins, but your stride may still be unique enough to identify you at a distance -- alone or among a group of people.
Researchers at the Georgia Institute of Technology and elsewhere are developing technologies to recognize a person's walk, or gait. Results indicate these new identification methods hold promise as tools in the war on terrorism and in medical diagnosis.
Gait recognition technology is a biometric method – that is, a unique biological or behavioral identification characteristic, such as a fingerprint or a face. Though still in its infancy, the technology is growing in significance because of federal studies, such as the Georgia Tech projects, funded by the federal Defense Advanced Research Projects Agency (DARPA).
At Georgia Tech, one study is addressing issues of gait recognition by computer vision, and the other is exploring a novel approach -- gait recognition with a radar system similar to those used by police officers to catch speeders.
The ultimate goal is detect, classify and identify humans at distances up to 500 feet away under day or night, all-weather conditions. Such capabilities will enhance the protection of U.S. forces and facilities from terrorist attacks, according to DARPA officials.
"We need technology to find the bad guys at a distance around federal buildings," says Jon Geisheimer, a research engineer at the Georgia Tech Research Institute (GTRI). "That is the original application. And after Sept. 11, we began to see the usefulness of these technologies in airports."
Because gait recognition technology is so new, researchers are assessing the uniqueness of gait and methods by which it can be evaluated.
"We know that we can get some information on gait, but that it is much less diagnostic than faces," says Aaron Bobick, an associate professor of computing and co-director of the computer vision project at Georgia Tech. " Currently, we can't recognize one in 100,000 people. At the moment, gait recognition is not capable of that, but it's getting better so it can act as a filter."
In its early development, gait recognition technology likely will serve as a screening tool in conjunction with other biometric methods.
With two years of experiments and analysis almost complete, researchers on both Georgia Tech projects are hopeful for continued funding to conduct further studies. They must address numerous technical issues and it will be at least five years before the technologies are commercialized, researchers say.
In the project using radar for gait recognition, results from experiments, data analysis and algorithm design are promising, says Geisheimer, who works under the direction of GTRI principal research scientist Gene Greneker, and collaborates with GTRI research engineer Bill Marshall and Georgia State University Professor of Biomechanics Ben Johnson.
Gait recognition by radar focuses on the gait cycle formed by the movements of a person's various body parts move over time.
"The magic goal we're shooting for is accuracy in the high 90 percent range," Geisheimer says. "We're not there yet, but our initial results are encouraging and promising."
Researchers correctly identified 80 to 95 percent of individual subjects, with variances in that range among the three experiment days.
The next step is to build a more powerful radar system and test it in the lab and then the field. In experiments last year, subjects started walking 50 feet away from the radar and then walked within 15 feet of it. But researchers are now building a radar system that can detect people from 500 or more feet.
In the study of gait recognition by computer vision, researchers distinguish their approach from others with a technique called an activity-specific static biometric. A static property – for example, a person's leg length -- is not a property of motion itself. It can be measured from a single image.
"The advantage of measuring a static property is that it is amenable to being done from multiple viewpoints," Bobick says. "…. Static measurements are view invariant, and that is a tremendous advantage because you can't control where someone goes."
Researchers are also developing statistical analysis tools for using a small, easily gathered database to predict how well a particular biometric, including gait recognition, will work on a larger population. These techniques will also help researchers determine what gait recognition properties to measure based on how well the technology can measure them.
"You can work on your ability to improve measurements," Bobick says. "But if you're not measuring something that is diagnostic, there is no amount of technology that will solve that problem with the biometric." Still in its infancy, computer vision-based gait recognition technology holds promise, particularly for verification or screening around the perimeters of government buildings or in an airport, if it is used in conjunction with other biometric technologies and information, Bobick predicts.
Meanwhile, researchers are applying the findings from their DARPA study to ongoing research in understanding human movement through video. Associate Professor Irfan Essa envisions applications in medical diagnosis. "Gait analysis has been important in the health field for a long time," Essa says. "Basic changes in someone's walking pattern can be an early indicator of the onset of Parkinson's disease, multiple sclerosis and normal pressure hydrocephalus (NPH)."
Further in the future are applications in diagnosing depression and lie detection.
Cite This Page: