Q: I know all about those facial-recognition cameras that are supposed to pick terrorists out of crowds. But what’s up with “gait recognition technology”? I was just reading how the Man’s got these new devices that can identify people based on their walking styles. Believe it or not?
The Pentagons’s pouring millions into gait recognition, so Mr. Roboto’s going with a tepid “believe it” for now. Top projects report success rates in the 90 percent range, but don’t worry yet about getting the bum’s rush because your bowlegged amble resembles that of an Al Qaeda fugitive. Despite the enthusiasm of Homeland Security pooh-bahs, this creepy brand of surveillance is still too raw for mass deployment.
Identifying no-goodniks by gait is nothing new, of course. Just this March, suspected serial killer Larme Price was nabbed, in part, because a detective recognized his “pigeon-toed gait.” (It helped that an obviously nervous Price walked right into the Brooklyn precinct, pretending to be an informant.) But the human eye can only pick out the most exaggerated walks; the theory behind gait recognition is that every saunter is somehow unique, and that only a machine can tell the differences.
This is the kind of technology that gets lips smacking at the Defense Advanced Research Projects Agency (DARPA), famous for sponsoring the weird and wondrous. (Check their goodies roster at darpa.mil.) And the DARPA folks most interested in gait recognition belong to the spookily named Information Awareness Office, headed by Iran-Contra retread John Poindexter. They’re in the midst of a $50 million project called “Human ID at a Distance,” which counts gait surveillance among its primary goals. Poindexter’s thus been giving away much cake to researchers at Carnegie Mellon, Georgia Tech, and a few other egghead campuses.
Carnegie Mellon’s approach (hid.ri.cmu.edu) involves filming the subject, then analyzing the swing and slope of each body part. It’s not a particularly novel idea, as athletics nerds have been using stop-motion video analysis since the heyday of pump sneakers. The CMU solution ups the ante by storing the “control” images, and then automatically comparing every target’s motion to the digital version.
The Georgia Tech alternative, more promising at the moment, uses a radar gun to measure the frequencies that bounce off ankles, legs, and feet. In mid-stride, for example, your knee’ll be a lot closer than your foot. (Remember the Doppler effect from high school physics? There you go.) The Tech team claims to correctly ID subjects 80 to 95 percent of the time, albeit in controlled circumstances. And it’ll work fine in inclement weather, as opposed to CMU’s motion-capture cameras.
DARPA’s looking to get a workable gait recognition system up and running sometime next year, one that’ll be able to spot someone from 500 feet away. The agency also wants to “fuse face and gait recognition into a 24/7 human identification system.” That sort of thinking gets the privacy community bug-eyed. Last October, the National Consumer Coalition’s Privacy Group (nccprivacy.org) named gait surveillance a “Villain of the Week,” carping on its accuracy statistics as overblown. More importantly, as alert author James Plummer pointed out, gait is not a constant like DNA. Try it out yourself—how hard is it to affect a little limp, walk on your insteps a smidgen more, or simply don a heavy overcoat? (Yes, that last strategy works.) And what if a basketball injury suddenly gives you the gait of a wanted Iraqi biochemist?
The hope here is that DARPA’s hip to gait recognition’s many flaws. If Poindexter and company plow ahead regardless, however, best to start perfecting your silly walk now. You never know when they might be on the lookout for your mosey. Especially you naughty subversives among the Mr. Roboto family.
This column marks Mr. Roboto’s one-year anniversary in the Voice. How ’bout a little feedback on what we’re doing wrong, and what we’re doing right? More Mac stuff, or less? More do-it-yourself guides, or less? More paranoia, or less? Please advise, treasured readers. Robots crave direction from their human masters.
Input questions at firstname.lastname@example.org.