SKDMap-Net debuts: security cameras could identify you by the way you walk

SKDMap-Net has been introduced in the International Journal of Reasoning-based Intelligent Systems as a new gait-recognition model. It identifies people by how they walk, not by their face, fingerprint, or iris.

The paper treats gait as a behavioral biometric, following joint bending , limb movement, stride timing, and rotation speed over time. On the numbers, it reports 95.8% accuracy on a major dataset and 83.7% Rank-1 accuracy on a tougher real-world benchmark. The model is meant to keep working when a face is turned away or covered, when parts of the body are blocked from view, and when the subject is being recorded from a distance in places such as streets, transit hubs, and campuses.

If you track surveillance and biometric systems, this one deserves a look. Gait can be captured remotely, with some reports putting the range at up to 50 meters. Other researchers and companies, including Watrix, have also published very high results. Real-world performance can still drop, though, because of lighting, clothing, injuries, bags, crowds, and camera angle.

The paper is available in the journal. Expect the privacy argument around this kind of tech to get louder as regulators in the UK and New York weigh whether gait should count as personal identifying information. That debate is picking up even as the broader market, including healthcare uses, is projected to grow from $1.25 billion in 2024 to $3.41 billion by 2032.