After 31 million tests, AI camouflage patterns beat every camera tried
A security researcher in Kansas City spent a year running the same test over and over, hoping to build something that could quietly outsmart the cameras watching American streets. The result is a set of AI camouflage patterns that, once printed on clothing or wrapped around a car, appear to stop surveillance software from recognizing what it’s looking at. Bill Swearingen calls the project noRecognition, and it just had its first public test at Def Con in Las Vegas.
Key takeaways
- Bill Swearingen’s noRecognition project uses AI camouflage patterns designed to stop surveillance camera software from identifying people, faces, or vehicles.
- The patterns defeated all 11 open-source detection algorithms Swearingen tested, including software tied to Flock license plate readers, Axon body cameras, and Clearview AI.
- It took roughly 31 million test runs to train the model to generate reliably effective patterns.
- The first public demonstration happened Friday at Def Con, where a wrapped 2009 Toyota Yaris rolled past a Flock camera undetected.
- The patterns cover vehicle bodywork rather than license plates, a deliberate choice meant to sidestep plate-obstruction laws.
AI-Driven Camouflage to Evade Surveillance Cameras
noRecognition works by exploiting a gap between human vision and machine vision — a wrap that looks like loud graphic design to a person can register as nothing at all to a detection algorithm. That gap is the whole premise behind Swearingen’s project, and it’s why the patterns matter beyond a single roadside stunt.
… Continue reading the full article at the original source below.
