Behavioural Biometrics in Virtual Reality: To What Extent Can We Identify a Person Based Solely on How They Watch 360-Degree Videos?

Recent years have seen a growing number of research papers that address the possibility of identifying users in virtual reality (VR) based on behavioural data. The main thrust of these considerations is applied research towards designing elaborate VR user authentication methods that are resistant to “shoulder surfing”. This widespread approach relies heavily on requiring users to perform elaborate tasks, and fails to explore the more fundamental scientific question of whether users can be identified under conditions of unforced behaviour-regardless of whether the purpose of such identification is to aid authentication, or, conversely, to violate user privacy. In our study
$(N=43)$
, we focused on examining the extent to which it is possible to identify the viewer of a previously unseen 360-degree VR video based on eye-tracking and VR headset orientation data that had been collected while the same viewer watched other 360-degree VR videos. We reached out to deep learning and classical statistical methodology and obtained a maximum average identification accuracy of 75%. Our results indicate that it is worthwhile to continue research efforts towards investigating the possibility of VR user identification under conditions of free, unforced behaviour.