Chinese Government Uses “Gait Recognition” Technology To Identify People

They say that you by observing the way people walk, you can tell a lot about them. A China-based startup has developed a technology allowing users to identify individuals based on their gait even when their face is hidden, or their back is facing the camera.

This technology called gait recognition is invented by Watrix, an AI startup. To build this software, the company needs to analyses everything about how a person walks, like their posture, body shape, arm movement, etc.

These traits will be put into a database that whenever the software wants to identify people, it needs to go through. According to Watrix’s official statement, the accuracy rate of gait recognition technology inside the lab exceeds 96%.

Although facial recognition technology has been evolving in recent years, there is still one critical weakness. The technology needs to view the people’s faces for identification. Therefore, if a person’s face is hidden or they have their back facing the camera, it’s useless.

On the other hand, gait recognition is more reliable. It can work basically from any angle. Since it analyses many different details, it is able to identify subjects even when they try to hide their legs. Fake limps or disguises will not work.

Watrix’s co-founder and CEO, Huang Yongzhen, said during an interview:

In October 2018, Watrix unveiled the very first version of the company’s gait recognition technology. Since then, it has been in discussion with several security companies from the Czech Republic, the Netherlands, Russia, India, and Singapore for contracts. Public security officials have also tested the technology for hours. The technology has also been used in police cases.

Authorities in Beijing and Shanghai are already utilizing gait recognition for wanted criminals detection and jaywalkers identification.

Although the current version of the software isn’t able to detect subjects in real-time yet, the Chinese firm claims that it is capable of analyzing a 1-hour video in 10 minutes, with more than 94% accuracy.