Project noRecognition: Teaching AI to Fool Surveillance Cameras
Patterns developed after 31 million tests successfully bypass facial recognition systems.
1The Concept of Evading Surveillance
Cybersecurity researcher Bill Swearingen from Kansas City spent a year developing patterns that surveillance cameras can't recognize. The noRecognition project doesn't prevent cameras from recording but disrupts the software layer (license plate recognition, facial tracking, etc.). The core philosophy isn't hiding from lenses but making algorithms 'indifferent' to what they see.
2The 31 Million Trial Process
The project began with Swearingen's concern about being tracked during a protest. What started as a simple experiment evolved into a reinforcement learning system. Through 31 million tests, the model learned to create patterns, learn from failures, and continuously improve. After each failed attempt, the system adjusted itself, eventually learning to bypass multiple detection algorithms simultaneously.
3The Gap Between Digital and Physical Tests
The research dashboard at sandbox.norecognition.org shows a 61.7% non-detection rate against a detector taken from an actual surveillance camera. However, these results are based on digital simulation. Swearingen's physical test on a real 2009 Toyota Yaris at DEF CON highlighted the critical difference between simulation and real-world performance.
4Real-World Test Results
At the DEF CON cybersecurity conference in Las Vegas, Swearingen tested one of his newest patterns on a car against a Flock Safety camera. While generally successful, the vehicle's wheels emerged as a weak point. Curved surfaces proved less compatible with flat printed patterns compared to flat car doors.
5Commercial Application Plans
Swearingen isn't publishing his best patterns to prevent camera manufacturers from easily blocking them. Instead, he plans to develop printed products like t-shirts and hoodies through crowdfunding. Vehicle wraps might come in later phases. The approach balances innovation protection with potential commercialization.
6Criteria for Practical Use
It's still unclear whether the project will become a practical privacy tool for everyday users. Real effectiveness will depend on how well the patterns work on actual clothing under various weather and camera conditions. Performance in rain or snow hasn't been tested yet, representing a significant knowledge gap.
7Industry Position and Implications
noRecognition responds to growing privacy concerns amid widespread facial recognition technology. If successful, it could empower individuals to demand more privacy in public spaces. However, the legal status of such technologies remains unclear in many jurisdictions, creating potential regulatory hurdles.
8Next Steps and Limitations
Swearingen's biggest challenge is ensuring consistent performance across different camera systems and conditions. Additionally, the potential for malicious use raises ethical questions. The project's long-term success will depend on both technical robustness and societal acceptance, requiring careful navigation of complex issues.
Original article: Security Affairs — Yapay Zeka ile Gözetim Kameralarını Şaşırtma Projesi