Realtime Behavior-Based Continual Authentication of Users in Virtual Reality Environments

Abstract

In this work, we present a realtime interactive system that uses the trajectories of the device controllers and headset of a virtual reality (VR) system to perform continual authentication of users interacting in VR. Our system identifies a user within .21 +/- .04 seconds of provision of a trajectory for a ball-throwing application. Given an interaction session authorized for a user within an organization, our system continually authenticates the current user by computing matches for the positions and orientations of the points on trajectories of the right hand controller, left hand controller, and headset for the user against controller and headset trajectories in a library of users. The system combines the position and orientation matches using a neural network pre-trained to predict confidences on matches computed within the library, and provides a speech interface to output the identified user as the library user with the maximum value of predicted confidence. The system uses the identity of the user authorized to use a session and a confidence threshold to detect if the user is genuine, a malicious user internal to the organization, or an external intruder. Our system can be readily ported into virtual reality applications that require realtime behavioral authentication of users for secure access.