In 2001, the Tampa, Fla., police installed a video camera with face recognition on the streets of their city. This was after 9/11, when there was a lot of hype about face recognition technology and how it could be combined with video surveillance cameras to stop terrorism and other crime. We filed an open-records request, obtained the Tampa police log files for the system, and discovered that it was failing miserably. (A report we wrote on those files, Drawing a Blank, helped spread word that the technology was not up to snuff.
Apparently it still isn’t.
Last year the German federal police announced that they were setting up a trial deployment of face recognition on the stairs of busy Mainz railway station in western German, where they automatically observed 23,000 commuters daily.
Yesterday we learned that the police have abandoned the test because the technology simply wasn’t good enough. The system only correctly identified “wanted” passersby 30 percent to 60 percent of the time. Reports haven’t divulged the false-positive rate (the rate at which it incorrectly identified innocent passersby as “wanted” suspects), but it’s a fair assumption that it was too high. (More details available in this article in Der Spiegel, but it is written in German).
Some computer-recognition tasks have improved greatly in recent years, such as voice recogntion. But face recognition may never be good enough for this kind of application. The human brain is hard-wired to recognize faces, after all, and evolved for millions of years to do just that – but still makes errors and can still be fooled.