NEC recognition tech can spot people even when their faces are obscured
You might think you're safe from being spotted by a camera if your face is covered or turned away, but the latest developments in AI-powered recognition tech are coming for you. NEC has announced a system that can identify people who are partially obscured, or even facing the other way.
Using a combination of deep learning techniques and analysis of body shape and clothes, the developers behind the system say they've hit accuracy rates of up to 90 percent during testing. It's officially called Person Re-identification Technology.
This use of whole body recognition rather than just the face could be used by security cameras in areas covering crowds, or where there are a lot of visual obstructions. The technology might be helpful in tracking people moving from place to place or room to room, even if their faces aren't always fully visible, NEC says.
The system still requires a source image – a picture of the person to look for – but it can then effectively use what glimpses of a person might be visible to find a match. It works from numerous angles with multiple cameras, and even from behind.
This new enhancement builds on top of the video face recognition technologies that NEC is already developing: technologies that can identify faces with a high level of accuracy even when they're moving, not well illuminated, or only in a small part of the video frame.
The company is also working on a Safer Cities initiative designed to help large crowds of people stay safe and minimize congestion – think airport arrival and departure lines, entry at sports stadiums, sensitive government facilities, and so on.
As with any surveillance technology like this, there are privacy implications. While the thought of criminals being apprehended more quickly is a reassuring one, no one likes to feel they're being watched without good reason.
The question of how this technology should be deployed will continue to be debated, but there's no doubt that it's getting smarter all the time.