For years, facial-recognition technology has become more and more popular. Even just a few years ago, facial recognition sounded like pure science fiction – now it's being used to automatically tag your Facebook photos.
Of course, the technology isn't perfect (as Google can attest to), and there's one thing that can thwart just about any modern facial-recognition software: bad lighting. At this point, most modern facial-recognition systems need clean, well-lit photos to function properly, although that might not be the case for much longer.
As part of a new study, researchers from both the Institute of Anthropomatics and Robotics and the Karlsruhe Institute of Technology have begun developing a new type of facial-recognition system – and it'll be a lot harder to fool than Facebook or Google Images. That's because the system doesn't rely on traditional methods. Instead, it focuses on a person's thermal signature to determine who they are. Instead of requiring light to function, this new system could function even in complete darkness.
At first, that might sound like it doesn't make any sense – a person's thermal signature can be changed in a variety of ways, and even simply walking outside can change how an infrared signature is read. For example:
If something as simple as a change of expression can change how a thermal image is interpreted, how can this system ever provide accurate results?
Well, by training a "deep neural network" – a computer system that's designed to mimic how the human brain works – the system can identify a person's face by making connections between different sets of information. Basically, by showing a computer enough info (in this case, normal photos), it can then transfer that information to an infrared image and make a connection between the two.
Of course, the system isn't quite perfect just yet. Even when provided with a number of different well-lit images, the system was only accurate 80 percent of the time, and was only successful 55 percent of the time when presented with a single normal image. It's clear that the technology is still in its infancy, but considering just how many photos of any given person are online nowadays, there's really no shortage of information available.
On top of that, the technology's already proving more accurate than other, similar systems. The study notes that the thermal imaging recognition system was already 10 percent more effective than other state-of-the-art tech – it's a big gain, especially for a system that's still so young.
If the technology continues to improve, it could prove to be a huge asset for law enforcement and surveillance teams. Again, there's no shortage of photos anywhere on the Internet, and that doesn't even include government databases.
Who knows – sooner or later, you might not even be able to hide in the dark.
Photo: Nadja Peek | Flickr
