Select your localized edition:

Close ×

More Ways to Connect

Discover one of our 28 local entrepreneurial communities »

Be the first to know as we launch in new countries and markets around the globe.

Interested in bringing MIT Technology Review to your local market?

MIT Technology ReviewMIT Technology Review - logo

 

Unsupported browser: Your browser does not meet modern web standards. See how it scores »

{ action.text }

Mark Nixon may be one of the pioneers of gait-recognition technology, but he credits Shakespeare with the idea: “Great Juno comes; I know her by her gait,” cries Ceres in The Tempest.

“You can find it in literature,” says Nixon, a computer scientist at University of Southampton in the U.K. “You can also find it in psychology and in medicine. Now we want to know if computers can do it.”

Interest in surveillance and recognition technologies since the September 11 attacks has thrust the spotlight on the U.S. Defense Advanced Research Projects Agency’s (DARPA’s) two-year-old $50-million Human ID at a Distance program. And while automated face recognition (See “Recognizing the Enemy” Dec 2001) receives the most attention, DARPA is also funding efforts at a handful of universities to identify people through their body language. The theory is simple: in the same way that each person has a unique signature or fingerprint, each person also has a unique walk. The trick is to take this body language and translate it into numbers that a computer can recognize.

One approach is to create a “movement signature” for each person. Researchers at Carnegie Mellon University’s Robotics Institute begin by filming individuals walking and running on a treadmill. Analog cameras tethered to a computer capture and store the action. Software tools remove any background footage, leaving a series of silhouettes of each subject, which are then stored as digital images. The same people are filmed again in an entirely different context, and the computer is instructed to identify the person against the stored images. “The system generalizes well across all the different gaits,” says Robert Collins, a research scientist at the Robotics Institute. “So far we’re getting a 90 to 95 percent correct match.”

A team at Georgia Tech, led by computational-vision researcher Aaron Bobick, uses a method called structural analysis to measure properties like a person’s stride length and leg spread. “We’ve run experiments both indoors and outdoors where we have to deal with issues like shadows and sunlight, and so far we’ve done pretty well,” Bobick says.

Gait recognition researchers face many challenges. One of the toughest: so far all database images are two-dimensional and depend greatly on the angle of the camera. When a system tries to compare two shots of the same person, taken from different angle, it is far less effective. A team at MIT’s Artificial Intelligence Laboratory, led by computer scientist Trevor Darrell, confronts this problem with computer-graphics techniques that re-render images at new angles. For example, if the database contains data for a person filmed from straight ahead, and in later footage the person walks in a zigzag motion, the graphics software will recreate a crude geometric model of the straight-on view, then compare it to the database. “It explicitly re-visualizes the image as if it was a straight line, and then runs the old algorithm,” says Darrell. In tests to identify 30 people, he says, “We’re still in the same ballpark ofroughly 95 percent accuracy.”

Although DARPA’s interest is primarily in potential military and security applications, researchers envision a broad range of uses. CMU’s Collins imagines integrating gait recognition into department stores or supermarkets, not for security but for marketing. Gait recognition might identify the shopping patterns of different demographics, or recognize shoppers who return within a half-hourthen forget them to protect their anonymity. Darrell imagines it used with every-day computer interfaces. “If I have a device that knows whether or not I’m looking at it, that can be useful, especially if the computer is trying to send me a message,” he says.

Still, this is a technology in its infancy, and no one is willing to predict when, if ever, to expect commercial applications. “We really don’t know yet how discriminative a person’s gait is,” says Darrell. “We don’t yet have any proof of concept that you can get a one-in-a-billion discrimination.”

Obviously a lot of ground needs to be covered before anyone aims for that one-in-a-billion ID. Says Nixon, “For now, it looks encouraging, but before we go to application, we have to ask, what support is there for gait as a basic biometric?”

As the Bard would say, that is the question.

0 comments about this story. Start the discussion »

Tagged: Communications

Reprints and Permissions | Send feedback to the editor

From the Archives

Close

Introducing MIT Technology Review Insider.

Already a Magazine subscriber?

You're automatically an Insider. It's easy to activate or upgrade your account.

Activate Your Account

Become an Insider

It's the new way to subscribe. Get even more of the tech news, research, and discoveries you crave.

Sign Up

Learn More

Find out why MIT Technology Review Insider is for you and explore your options.

Show Me