For all the giant exploding Death Stars in SciFi, its really the mundane devices that stay with us for years after. Doctor Who‘s sonic screwdriver, Picard’s replicator, and Spock’s tricorder have at least as much resonance for us as any gigantic space laser that ever turned a plot. In Knight Rider, our resident crime fighters rely pretty heavily on KITT’s ability to find people. He accesses a government database — usually the DMV — and then connects to various surveillance cameras in the area (Knight Rider crooks do tend to like Vegas casinos). The ability to access closed-circuit cameras aside, what’s really amazing here is KITT’s ability to digitally match photos to a moving image. For modern law enforcement and software search companies, that’s something of a holy grail.

For Google, really good facial recognition software would kick down the barriers to searching images online. Right now the company’s Image Search function relies on file names and captions supplied by users to make a match. That’s why a search on Michael Knight captures both pictures of the 80′s TV star David Hasselhoff and shots of clothing designed by Project Runway winner Michael Knight. The closest Google has come to solving the problem is a tool to enable users of their Picasa Web Albums to organize their pictures better. Users identify people in some photos, and the system creates a profile of that person, taking into account data like the depth of a person’s eye sockets or the distance between the eyes and the nose, or the eyes and each other. The software can then sift through pictures and look for additional photos of that person, saving the user from perpetually keeping their photos organized (which, frankly, none of us do anyway, right?). Google claims the function is fairly accurate, though shadowing or washed out photos can cause it problems.

But the real heavyweights behind facial recognition technology are the cops and private security firms. Remember in Total Recall, when Ahnold runs through the full body scanner? The scanner there both searched for weapons and took a scan of his whole body compared the images to a data base. Contemporary law enforcement has more modest goals, like matching a simple two dimensional picture to the face of someone caught by a video camera or walking through a security scanner. So far, they haven’t had much luck: In 2001, Tampa police tried to use it in areas with a buzzing night life, and in 2002, security in Boston’s Logan airport tried to use it. But in both cases, the recognition rate was so poor, they had to scrap the systems. The British government is in the midst of a face-scanning debacle of its own at Manchester Airport. The system, still in the midst of a six month test, not only fails to make the best matches, but if two people walk through the scanner together, it will only notice the first one.

But these are all application problems. In 2006, computer scientists competing in the Face Recognition Grand Challenge discovered that their algorithms were so good that, under ideal conditions, they could pick out faces better than even people do. The software works so well that it can ID faces turned as much as 90 degrees. What KITT has mastered, and what we’re still working on, is applying that technology here in the real world with imperfect light, moving targets, and the evasions of people who maybe don’t want to be recognized. Seems like we’ll get there sooner or later.

October 15th, 2008 Tags: David Hasslehoff, face recognition, google, Knight Rider, Michael Knight
by Eric Wolff in TV, Uncategorized | 5 comments | RSS feed | Trackback >