Crypto
Home›Crypto›Market Structure›Adversarial camouflage pattern defeats open-source cam…
Adversarial camouflage pattern defeats open-source camera AI detection
A Kansas City researcher says he ran 31 million tests to generate patterns that prevent object-detection software from logging vehicles, including in a Def Con demo with a Flock camera.
A Kansas City security researcher says an AI-generated camouflage pattern can prevent camera software from correctly classifying what it covers, including people, faces, and cars, by disrupting object-detection models rather than “blinding” the underlying cameras.
According to Decrypt, the researcher, Bill Swearingen, tested his noRecognition project against 11 open-source detection algorithms, and said it defeated all of them, including systems associated with Flock license plate readers, Axon body cameras, and Clearview AI. The first public demonstration took place Friday at Def Con in Las Vegas, using a 2009 Toyota Yaris wrapped in the pattern and driven past a Flock camera.
Decrypt reports that the pattern does not stop the camera from recording, with footage still captured normally for a human viewer. What changes is the AI layer, the object-detection model that decides whether to log a vehicle or plate, which Swearingen says records nothing because the pattern creates engineered visual noise.
Swearingen said he built the system using reinforcement learning, where the model effectively grades and adjusts its own output, and he claims it can generate fresh patterns every minute while keeping the strongest versions offline so camera vendors cannot train against them. He also said the concept came from wanting to attend a protest without being tracked, and he described the work as a way to opt out of being logged by surveillance systems.