A First Look into Privacy Leakage in 3D Mixed Reality Data

Abstract

We have seen a rise in mixed (MR) and augmented reality (AR) applications and devices in recent years. Subsequently, we have become familiar with the sensing power of these applications and devices, and we are only starting to realize the nascent risks that these technology puts over our privacy and security. Current privacy protection measures are primarily aimed towards known and well-utilised data types (i.e. location, on-line activity, biometric, and so on) while a few works have focused on looking into the security and privacy risks of and providing protection on MR data, particularly on 3D MR data. In this work, we primarily reveal the privacy leakage from released 3D MR data and how the leakage persist even after implementing spatial generalizations and abstractions. Firstly, we formalize the spatial privacy problem in 3D mixed reality data as well as the adversary model. Then, we demonstrate through an inference model how adversaries can identify 3D spaces and, potentially, infer more spatial information. Moreover, we also demonstrate how compact 3D MR Data can be in terms of memory usage which allows adversaries to create lightweight 3D inference models of user spaces.