Towards Protecting 3D Point Cloud Data in XR Data Sharing through Perceptual Signatures

As Extended Reality devices with advanced spatial sensing become more common, protecting user privacy during 3D data-sharing is increasingly important. This paper lays the groundwork for a privacy-preserving framework that automatically detects and removes sensitive objects from 3D spatial maps before sharing them with third-party applications on the device. The core component of our framework is a perceptual signature generation method that uses Locality-Sensitive Hashing to create secure, compact, geometry-based signatures for 3D object point clouds. These signatures allow for fast and reliable identification of private objects, even when 3D spatial maps are captured under different resolutions, or noise. We first evaluate our method on synthetic data built from ModelNet40, then show that it also performs well on real-world data captured using Meta Aria glasses. The results confirm that our framework can effectively protect sensitive objects, validating our approach towards a practical solution for secure 3D spatial data-sharing in everyday XR applications.