Within-System and Cross-System Behavior-Based Biometric Authentication in Virtual Reality

In this paper, we perform the first analysis of behavior-based authentication within and across multiple virtual reality (VR) systems. The proliferation of VR in everyday applications such as healthcare, manufacturing, education, and remote teleoperation necessitates an understanding of operation over multiple VR systems as it is likely that a user will utilize several systems and requiring the user to re-train to a new system is time consuming. We collect the first multi-system dataset for VR biometrics consisting of 46 users performing a ball-throwing interaction using the Oculus Quest, HTC Vive, and HTC Vive Cosmos. Each user provides 10 training samples on the first day and 10 test samples on the second day for each of the three devices. We collect position and orientation data for devices such as the headset, left hand controller, and right hand controller, as well as trigger positions for the dominant hand controller. Our approach uses pairwise matches between trajectory features to represent high intra-user consistency and inter-user discriminative capacity. We analyze within-system and cross-system authentication accuracy for the 41 right-handed users in our 46-user dataset over varying combinations of matches for features such as point positions, device orientations, linear and angular velocities, and trigger grab or release for the right hand controller, left hand controller, and headset. Our approach provides maximum within-system authentication accuracy of 97%, 91%, and 91% when test trajectories are compared to training trajectories for the Vive, Quest, and Cosmos respectively. We provide maximum accuracies of 58% for testing with Cosmos and training with Quest, 70% for testing with Cosmos and training with Vive, and 85% for testing with Quest and training with Vive.