Infrared Light Attack Targets Optical Flow Networks in Real Time
A novel technique has been created by researchers for conducting real-time physical attacks on Optical Flow Estimation Networks (OFENs) utilizing infrared light. These networks play a vital role in autonomous driving and motion detection, providing essential outputs for subsequent processes. This new method involves the dynamic display of pre-generated adversarial examples (AEs) through infrared light, allowing for accurate and targeted assaults without altering the target system. In contrast to earlier digital-to-physical methods, this approach directly impacts the model in the real world. The findings are documented on arXiv (2607.26651) and underscore the weaknesses in AI systems dependent on optical flow estimation.
Key facts
- Attack targets Optical Flow Estimation Networks (OFENs).
- Uses infrared lights for stealthy, real-time attacks.
- Pre-generates adversarial examples and displays them dynamically.
- Enables precise, targeted attacks without modifying the victim system.
- Directly attacks the victim model in the physical world.
- OFENs are used in autonomous driving and motion detection.
- Published on arXiv with ID 2607.26651.
- Method differs from previous digital-to-physical attack techniques.
Entities
Institutions
- arXiv