Priv360: Application-Oriented QoE-Optimized Client-Side Protection for 360-Viewer Identification
Authors: Sheyda Mirzakhani (Linköping University), Niklas Carlsson (Linköping University)
Volume: 2026
Issue: 4
Pages: 1106–1123
DOI: https://doi.org/10.56553/popets-2026-0162
Abstract: Head movement data in virtual reality (VR), particularly during 360° video streaming, can reveal uniquely identifying behavioral patterns, posing serious privacy risks. While noise injection can obscure these signals, it often degrades the user’s Quality of Experience (QoE), creating a challenging privacy–utility tradeoff. We introduce Priv360, a client-side defense that injects carefully tuned noise into the transmitted 6-DoF head pose while preserving the user’s actual viewing experience. Before sending metadata to the server, the client reconstructs a stable predicted viewport from the noisy pose using a fast AR(2) model fused with a constant-jerk Kalman filter (with an optional LSTM-enhanced variant). Only this predicted viewport is transmitted for quality adaptation; the client continues to render the true viewport locally using the unperturbed pose. Using real 6-DoF datasets, including a newly collected Meta Quest~3 dataset, and a leave-one-video-out evaluation, we show that Priv360 substantially reduces re-identification accuracy while maintaining high visual quality across noise levels, prediction horizons, bandwidth settings, and attacker architectures. We further compare multiple prediction filters and show that combining learned and model-based predictors yields the most favorable privacy–QoE tradeoff. Our results provide a practical privacy defense and actionable insights for deploying privacy-aware VR streaming.
Keywords: Privacy, PET, Client-side protection, QoE, VR, 360 video
Copyright in PoPETs articles are held by their authors. This article is published under a Creative Commons Attribution 4.0 license.
