Knowledge-driven Biometric Authentication in Virtual Reality
Florian Mathis, Hassan Ismail Fawaz, Mohamed Khamis
Abstract
With the increasing adoption of virtual reality (VR) in public spaces, protecting users from observation attacks is becoming essential to prevent attackers from accessing context-sensitive data or performing malicious payment transactions in VR. In this work, we propose RubikBiom, a knowledge-driven behavioural biometric authentication scheme for authentication in VR. We show that hand movement patterns performed during interactions with a knowledge-based authentication scheme (e.g., when entering a PIN) can be leveraged to establish an additional security layer. Based on a dataset gathered in a lab study with 23 participants, we show that knowledge-driven behavioural biometric authentication increases security in an unobtrusive way. We achieve an accuracy of up to 98.91% by applying a Fully Convolutional Network (FCN) on 32 authentications per subject. Our results pave the way for further investigations towards knowledge-driven behavioural biometric authentication in VR.