EmotionPass: Unobtrusive Emotion-based User Authentication on Glasses
Ying Hao, Chengzhang Yu, Noman Ahmed, Yincheng Jin, Zhanpeng Jin · 2024
Wearable devices are increasingly prevalent, necessitating reliable user authentication to mitigate privacy risks. Glasses, as a prominent category of wearable devices, have traditionally employed methods such as iris recognition, depth cameras, and electrooculography (EOG) for authentication purposes. However, these solutions are usually only applicable for bulky VR/XR goggles and are often impractical for emerging lightweight smart glasses due to the need for additional hardware modules. Thanks to their compact size, low cost, and privacy-preserving capabilities by capturing muscle movements, Inertial Measurement Units (IMUs) offer a promising alternative for user authentication. This study introduces EmotionPass, a novel system embedding IMU sensors in a pair of lightweight glasses for passive user authentication via facial expressions. Utilizing a Knowledge-Aware Network (KAN) and Supervised Contrastive Learning, our system extracts distinctive features from IMU signals. Experiments with 23 participants performing seven facial expressions demonstrated a high average accuracy of $99.22 \%$ and an Equal Error Rate (EER) of 0.0154 in user attack scenarios. We compared various baseline algorithms and proved the reliability and stability of our algorithm. The system maintained $92.51 \%$ accuracy for unseen expressions, with long-term usability confirmed. EmotionPass thus offers a non-invasive, cost-effective, and efficient solution for continuous user authentication in wearable settings.