Researcher Bill Swearingen demonstrated the noRecognition pattern at the Def Con conference, which prevents smart camera algorithms from recognizing vehicles and people in video footage, reports infohub.kz.
At the Def Con cybersecurity conference in Las Vegas, researcher Bill Swearingen presented the noRecognition project. In the experiment, a 2009 Toyota Yaris was wrapped in a film with a special mathematical pattern and driven past a Flock camera. The automatic recognition system failed to identify the object as a vehicle.
The development uses adversarial machine learning techniques. The pattern does not cover the lens or obstruct video recording, but it confuses computer vision algorithms. As a result, the video is still recorded, but the automation does not flag the object or enter it into databases, keeping the person or car anonymous.
Swearingen trained the model using reinforcement learning over a year in Kansas City, conducting 31 million tests. The graphic pattern managed to fool 11 popular recognition algorithms, including software for Flock cameras, Axon body cameras, and Clearview AI systems. The author plans to produce clothing and vinyl wraps for cars with protective prints.