Identifying People by the Way They Walk

Michael Jackson has his infamous moonwalk, John Wayne had that memorable cowboy swagger, and Mae West was noted for her remarkable sashay.

But according to scientists, one day it may be possible to recognize anyone by the way they walk, or their gait.

Several universities have been working with the Defense Advanced Research Projects Agency to develop gait recognition as another viable biometric tool.

It's hoped that if gaits are as unique as say voice patterns or fingerprints, advanced surveillance systems could identify and track people by the way they walk through sensitive areas such as airports or around government buildings.

From the Way That You Walk

To study gait, most of the various research projects use video cameras and computers to capture and create so-called "movement signatures."

When a person walks, body parts — the legs, knee joints, arms, elbows, and so on — create a particular repeating pattern as they each move through space. The video camera captures these points of movements and sends them to a computer for analysis. The computer notes the movements and establishes mathematical relationships for each point to create the "signature" patterns it needs to recognize for each individual.

Researchers at the Massachusetts Institute of Technology and Carnegie Mellon University have been testing such gait recognition as part of the DARPA project with impressive results so far.

These experimental setups use signatures captured from volunteers walking on treadmills. When the systems capture a test subject walking in front of a different background, researchers report the machines make a correct match 90 percent to 95 percent of the time.

But there are still difficulties in such systems.

The biggest challenge: Since actual video images must be captured and analyzed, such gait recognition systems could only be used in well-lit areas. What's more, such systems would have a very difficult time picking up distinct movement points of a person far away and wearing bulky, concealing clothing such as a heavy coat.

Searching for Striders by Radar

But scientists at the Georgia Institute of Technology in Atlanta, are taking a slightly different approach with their gait recognition system.

Instead of video cameras, their experiments involve radar guns — similar to the ones used by police to track speeding cars.

The system sends out a pulse of radar energy and receives the signals reflected off objects. As a person walks, the radar signals change because of the so-called "Doppler shift" effect. Objects approaching closer generate higher frequencies than objects moving away.

Georgia Tech's system is sensitive enough to recognize and capture the differing Doppler shifts during anyone's natural walking motions. As a person walks toward the detector, certain parts — the right foot, for example — will have a higher radar signatures while other parts — the right hand, say — will have a much lower signal at the exact same moment.

By recognizing these patterns, the system creates a signature that helps it distinguish a moving person from say a leaf blowing across the radar beams.

"We're not seeing an image of the person," says Gene Greneker, the principle research scientist with the project at Georgia Tech. "We're just seeing the Doppler [shift] off those body parts and identifying what we're looking at with the radar."

Unblinking Eye

Greneker also says that the radar system isn't hindered by clothing or inclement weather — making it ideal for outdoor security and surveillance systems.

Greneker and his team say that labs tests have shown their system can recognize a person by their gait 80 percent to 95 percent of the time. But the eventual goal is to improve the accuracy to the high-90 percentile range.

Operating range is another feature the Georgia Tech scientists hope to improve. Last year, the experiments were limited to subjects walking 15 feet to 50 feet from the detector. This year, Greneker hopes to soon test a system that will be able to spot and identify walkers from 500 feet and beyond.

The Steady Pace of More Research

Researchers believe that gait recognition systems won't be commercialized for years to come since there's still plenty of research that needs to be done.

For example, scientists still aren't sure just how individually unique everyone's stride may be.

"It's like asking if fingerprints are unique," says Bill Marshall, a research engineer also with the Georgia Tech team. "Mathematically, there's a small percentage of people — one in a billion? — that have the same fingerprints. So, we're not prepared to say that everyone has a unique gait."

And it isn't clear how "robust" gait recognitions may be overall. Researchers still need to figure out if imposters can fool the system by mimicking someone else's strut. And conversely, they would also need to design the systems to figure out how to match people who have only a temporary change in their gait — say, a slight limp from a twisted ankle.

Many researchers say that gait recognition most likely won't work as a stand-alone biometric security scheme. Instead, they figure it could be a useful supplement to other biometrics and help provide for a multi-layered security scheme.

"One of the problems with vision-based face recognition is they are computational intensive," says Marshal. "If you could use gait as filter to narrow down the number of choices to look at, it might make the other biometric technology more practical."

DARPA says that gait recognition system research will continue to be part of its multi-year Human ID at a Distance project. One of the program's goals is to field-test an automated identification system that combines gait and facial recognition biometric technology by fiscal year 2004.

Comments