Gait recognition in place of face recognition

5 March, 2019 12:00 AM printer

They say you can tell a lot about a person by the way they walk, but one Chinese startup has apparently developed technology that allows it to identify individuals by their gait, even if their face is covered or they have their back to the camera.

Known as “gait recognition”, the technology invented by artificial intelligence startup Watrix analyses thousands of metrics about a person’s walk, from their body shape and the angle of arm movement to their posture and whether they have a toe-in or toe-out gait. All these individual traits go into a database that the software then goes through when attempting to identify people. According to an official statement from Watrix, the accuracy rate of gait technology at the laboratory level exceeds 96 per cent.

Facial recognition technology has come a long way in recent years, but it still has a fundamental weakness – it needs to “see” people’s faces in order to identify them, so if someone’s face is covered or if their back is to the camera, it’s basically useless. Gait recognition, on the other hand, is much more reliable. It works from virtually any angle, and because it analyses so many metrics, it can identify subjects even if they attempt to cover their legs. Disguises or fake limps are not very effective in fooling the technology either. “Covering your legs would reduce the recognition score but we analyse all of a person’s body,” Huang Yongzhen, co-founder and chief executive of Watrix, said in an interview. “With facial recognition people need to look into a camera – cooperation is not needed for them to be recognised [by our technology].”

Watrix revealed the first version of their gait recognition technology in October of last year and has since been in talks with security companies from Singapore, India, Russia, the Netherlands and Czech Republic for contracts. The technology has also been tested by public security officials for thousands of hours and used to solve dozens of police cases. Authorities in Shanghai and Beijing are already using gait recognition to detect wanted criminals and identify jaywalkers.

The current version of gait recognition software isn’t yet capable of detecting people in real time, but the Chinese startup claims that it’s able to analyse an hour’s worth of video footage in just 10 minutes, with over 94% accuracy.

Watrix also develops facial recognition technology and recommends combining it with gait recognition to improve overall detection results. Yup, good luck hiding from that combo, I guess…