We’ve written about computer vision and facial recognition many times before, and recently one of the issues that has come up has been that all the cameras in the world are generating more visual data than people can watch. I wrote about Kitware’s intention to create “behavior recognition” in camera systems, and this “Questionable Observer Detector” seems like it would be a nice complement to it.
You’ve heard, no doubt, of security cameras that snap pictures of your face as you pass and compare them to a central database of smugglers, terrorists, and other baddies. The trouble is that unless these people are already known, the system can’t really detect them. So a team of computer scientists at Notre Dame decided to put together a way for a database to be built on the fly.
The camera would analyze footage, recording the time and place of every face in it. Later, while looking at different footage, it would compare the new tracking data to the old, and see if anybody has shown up twice. Someone passing by the same corner five times in a week would be normal, but someone coming and loitering in an airport ticketing area twice a week for a month might be something to look into, regardless of whether that person’s face is in any databases or connected with any crimes.
[youtube http://www.youtube.com/watch?v=EUHz_i5VAUM w=640]
It’s a bit scary, but so is the problem it’s meant to combat: the kind of criminal who only needs one chance to commit a terrible crime — like a suicide bomber. They’ll never be in any databases for obvious reasons, but have a habit of checking out targets a few times before they attack.
Sure, it’s also the kind of thing that a police state would love to have in their arsenal. But someone had to invent it sooner or later.