Summary

  • Bill Swearingen's noRecognition initiative creates designs that prevent camera systems from identifying covered objects, such as people or vehicles.
  • These patterns successfully bypassed all 11 open-source detection algorithms tested, including those used by Flock license plate readers and Axon body cameras.
  • The first public demonstration occurred at Def Con in Las Vegas, featuring a 2009 Toyota Yaris wrapped in the pattern and driven in front of a Flock camera.

Bill Swearingen has dedicated the past year to an extensive experiment from his Kansas City home, where he co-founded the SecKC security meetup. After conducting around 31 million tests, he claims to have developed patterns that can effectively obscure objects from the detection systems of Flock cameras, a controversial surveillance technology being deployed nationwide.

He publicly showcased his work for the first time at Def Con, collaborating with the YouTube channel Donut Media to wrap a 2009 Toyota Yaris in one of his latest designs and drive it past a Flock camera.

Myriad: How many days will Claude go down in August? Click to make your prediction.

“We demonstrated its effectiveness,” Swearingen told TechCrunch, although he noted that the vehicle's wheels posed a challenge. Donut Media plans to release video footage of the demonstration in the coming weeks.

The pattern does not obscure the camera's ability to record; the footage remains intact, and an observer would still see a car. However, it disrupts the object-detection model that typically identifies the vehicle and its license plate.

Essentially, by introducing enough visual complexity tailored to confuse the AI's algorithms, the system fails to log the vehicle, making it indistinguishable amid other traffic.

Image credit: Donut Media

This method relies on adversarial machine learning principles, highlighting the disparity between human visual perception and machine interpretation. A design that appears vibrant and eye-catching to a person may be rendered meaningless to an AI classifier.

Swearingen developed the patterns using a reinforcement learning model that self-evaluates its performance. If a pattern is detected, the model adapts and refines its output—an approach he likens to teaching the model “how to paint.” The system generates new patterns every minute, with the most effective ones kept offline to prevent camera manufacturers from training against them.

“Privacy is a fundamental right,” he asserted, describing these patterns as a means for individuals to “opt out of being tracked.” The concept originated last year when he sought to attend a protest and was concerned about surveillance cameras capturing the identities of attendees.

The Ongoing Battle Against Detection Technologies

For years, individuals have sought ways to evade detection systems, often utilizing rudimentary solutions. Activists in San Francisco have previously placed traffic cones on the hoods of Waymo and Cruise robotaxis to immobilize them, a tactic that required no technical expertise.

During immigration raids in Los Angeles last year, some protesters escalated their actions by setting fire to several Waymo vehicles. Masks and hats remain common among demonstrators to obscure their identities. Additionally, adversarial clothing has been marketed with patterns designed to confuse facial recognition systems, while anti-recognition glasses have emerged, albeit with limited evidence of their effectiveness.

What distinguishes Swearingen’s noRecognition project is its targeted approach. He tested the patterns specifically against widely used detection systems, particularly Flock, which has drawn significant criticism. The company is currently facing increasing scrutiny from lawmakers, with internal documents revealing plans to convert 350,000 Uber and Lyft dashcams into a mobile plate-scanning network.

These automated readers have already wrongfully detained innocent drivers, and immigrants and activists continue to be caught in ICE’s AI surveillance net. Lawmakers are also pressing Meta regarding facial recognition technologies in its smart glasses, leading privacy advocates to explore legal avenues to combat automatic detection systems.

Swearingen’s noRecognition project is currently running a crowdfunding campaign to finance initial merchandise, including T-shirts and hoodies, with plans for vehicle wraps to follow. He aims to achieve a resolution that is effective at a distance and visually appealing enough for people to wear.

Driving a vehicle wrapped in these patterns raises legal questions, as laws regarding license plate visibility differ by state. However, the patterns are designed to obscure the vehicle's body, not the plates themselves.

“Each failure enhances my model, leading to continuous improvements in the patterns,” Swearingen stated.