EV-GazeLock: A User Authentication System Based on Micro Eye Movement with Event Cameras

Off-the-shelf VR/AR head-mounted devices are increasingly integrating eye tracking as a novel human-computer interaction interface, enabling new applications in virtual environments. The distinct and reproducible patterns of eye movements among individuals present an opportunity for user identification, known as eye authentication. However, current eye authentication solutions face limitations due to the low temporal resolution of CMOS/CCD cameras, leading to accuracy issues in robust authentication. Moreover, CMOS cameras capturing detailed appearance around the eye region raise concerns about privacy leakage. To overcome these challenges, we propose leveraging event cameras for eye authentication. Event cameras encode per-pixel intensity changes as an asynchronous event stream with high temporal resolution (tens of microseconds), capable of capturing subtle identity-related characteristics from eye movements, termed micro-motion. This paper introduces EV-GazeLock, a novel authentication approach based on eye movements captured by event cameras. EV-GazeLock uses only event streams as input and extracts identity-related characteristics through a custom MicroFlow network designed for spatial-temporal feature extraction from micro-motion. This effectively distinguishes eye movements between different users while avoiding recording detailed eye appearance. Extensive evaluations on multiple datasets demonstrate that EV-GazeLock significantly outperforms several state-of-the-art gaze authentication or recognition methods. Ablation studies further validate the design efficacy of MicroFlow, reinforcing the robustness and effectiveness of the proposed approach.