This ‘adversarial’ pattern can prevent surveillance cameras from detecting you

A patterned 2009 Toyota Yaris recently drove past a Flock surveillance camera at DEF CON and, according to researcher Bill Swearingen, avoided the automated detection system. TechCrunch describes the event as the first public physical-camera test for the noRecognition project, though the result remains much narrower than actual invisibility. The camera still successfully recorded continuous footage of the vehicle, and the demonstration video and system logs have not been made public. TechCrunch stands as the sole detailed public account of the drive-by.

This distinction holds weight because the project's own research dashboard labels its published benchmarks as digital and simulated. Those tests evaluate rendered patterns against detector software—including weights extracted from deployed cameras—rather than measuring printed fabrics or vehicles under uncontrolled outdoor conditions. The Las Vegas test serves as an intriguing lead rather than a fully reproducible scientific result, requiring independent verification across varying conditions, distances, lighting, and speeds.

At Black Hat, Swearingen presented research exploring whether clothing patterns can fool facial recognition, followed by the Friday vehicle test conducted with help from Donut Media. While the researcher stated the expected alert failed to trigger, the demonstration leaves many operational variables unaddressed. Crucially, the camera did not stop recording; the failure occurred solely at the automated layer that flags objects and generates alerts. A missed machine detection can bury footage inside a massive archive, but human operators reviewing the timeline can still spot the vehicle.

To address traditional vulnerabilities like the segment-missing problem—where fabric folds or changing angles render fixed patches useless—designer Simon Weckert has introduced a conceptual garment collection titled Digital Camouflage. Manufactured in Latvia from a durable blend of recycled and standard polyester, these garments utilize a seamless adversarial texture generated by a specialized AI method known as TC-EGA. Applied via digital textile printing, the continuous pattern produces high-frequency visual noise and false features to disrupt computer vision systems.

Evaluating how these textures perform across different architectures reveals varied efficacy.