London Olympics to sport photometric stereo facial recognition technology

All eyes will be on London in 2012 when the city by the Thames hosts the Olympics. And just as China began its security preparations years ahead of its forthcoming 2008 games, city planners in London are already putting together what promises to be the most sophisticated security operation to date -- an effort made all the more critical due to the city's own recent and very real problems with terrorism. There are many things to look for in 2012 security-wise, but one thing that caught our eye was a new facial recognition software under development at Imperial College London in coordination with General Dynamic and Identity Solutions.

Of course, facial recognition technology is not new, but the Imperial College approach distinguishes itself by the use of photometric stereo, a technique that vastly improves the accuracy of facial mapping and identification by creating what the researchers call a "facial skin signature." How does it work? The photometric stereo uses a series of lights and a fixed camera to generate an image. "When you have light coming from an angle on to a rough surface it creates shadows," said professor Maria Petrou. "The position of the light determines where the shadow falls -- and more lights means more images. You can then combine the shadow information with software to infer the 3D shape of the object." By attaching a camera to a series of sequential flashes from multiple sources -- done so quickly that the delays are unnoticeable -- the images can be combined to produce a 3D image.

This is a superior approach to existing systems for a number of reasons. First, conventional stereo imaging uses images from a number of different cameras which must then be matched. Fair enough, but then software is needed to accurately match each pixel from each individual image -- an effort that can lead to inaccuracy and ambiguity, especially when pixels can match in different areas of the image. The photometric stereo approach relies on only one camera with a single angle, thereby sidestepping the pixel matching program, and because the image stays the same but the lighting source changes (due to the flashing), only the shadows differ. In addition, the multiple lights can also provide information about skin tone and color, something normal facial recognition software cannot accomplish.