New AI-generated patterns can hide people and cars from street cameras

Researcher Bill Swearingen unveiled AI-crafted designs that, when printed on clothing or vehicles, prevent common surveillance cameras and license-plate readers from recognizing the covered objects.

Bill Swearingen spent the past year testing more than 31 million pattern variations to develop a method that defeats the detection capabilities of widely deployed surveillance cameras. His reinforcement-learning model now generates new designs each minute, each iteration improving at evading eleven open-source detection algorithms, including those powering Flock license-plate readers, Axon body-worn cameras and Clearview AI systems.

At Def Con in Las Vegas, Swearingen, aided by Donut Media, covered a 2009 Toyota Yaris with one of his latest patterns, successfully preventing the car from being identified by a Flock camera, though wheel detection remained a challenge. The project, called noRecognition, aims to offer privacy-preserving apparel and vehicle wraps, with a crowdsourcing campaign to fund early merchandise. Swearingen intends to keep the strongest patterns off-line to avoid counter-development by camera manufacturers while continuing to refine the model.