Unravelling Spatial Privacy Risks of Mobile Mixed Reality Data
summary
Analyzes 3D maps from Microsoft HoloLens and Android ARCore and shows that shape-similarity metrics predict when spatial inference succeeds. Implements two attackers, NN-matcher (local-structure) and PointNetVLAD (neural), and varies revealed area to measure success. Finds distance-based chi_d2 correlates weakly or inversely, while normal-vector measures chi_v2/chi_xi correlate strongly with NN-matcher success; deep-learning attacker success shows no consistent correlation, so learned models bypass complexity. Increasing revealed space raises attacker success.
AI-written summary · about
related talks
53mA Thorough Look into Spatial Mapping with HoloLens
06mWhen Collaboration Betrays: The Evolving Attack Surface of...
15mA Closer Look: Evaluating Location Privacy Empirically
41mComputer Vision for Remote AR
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...
46mFrom Health Data to Location Traces: Addressing Spatiotemporal Correlations in Differential Privacy
23mAnalysis of Privacy Protections in Fitness Tracking Social Networks -or- You can run, but can you hide?
01h 43mPerspective