Adversarial Patterns Hide People From AI Vision Systems
TL;DR. A researcher demonstrated adversarial patterns at Defcon that significantly lowered AI camera detection confidence for human presence. - The project, noRecognition, aims to develop clothing that makes individuals harder for AI surveillance systems to detect. - This technology targets the initial person-detection layer, not facial identification, confusing computer vision. - Real-world effectiveness on worn clothing remains an unproven test, despite successful demonstrations.
- A researcher developed adversarial patterns to confuse AI person-detection systems.
- The patterns, shown at Defcon, reduced an AI camera's confidence score for detecting a human.
- The noRecognition project seeks to integrate these patterns into clothing for surveillance evasion.
- This research focuses on evading the initial AI layer that detects human presence, not facial recognition.
- The practical application of these patterns on worn clothing is still under development and testing.