Toward Multimodal Privacy in XR: Design and Evaluation of Composite Privatization Methods for Gaze and Body Tracking...

Area of Science:

• Extended Reality (XR) and Human-Computer Interaction
• Cybersecurity and Privacy Engineering
• Behavioral Biometrics

Background:

• Extended reality (XR) systems collect sensitive behavioral data, including eye and body telemetry.
• Existing privacy mechanisms often fail to prevent re-identification when applied independently to different data modalities.
• The increasing immersiveness of XR necessitates robust, real-time privacy solutions.

Purpose of the Study:

• To systematically evaluate the effectiveness of real-time privacy mechanisms for XR systems.
• To assess both individual and paired privacy mechanisms across eye and body data modalities.
• To balance privacy protection with the utility requirements for real-time XR interactions.

Main Methods:

• Evaluated four eye and ten body privacy mechanisms on multiple datasets with up to 407 participants.
• Assessed privacy using re-identification rates and utility via established performance thresholds.
• Tested mechanisms in both casual (e.g., VRChat) and competitive (e.g., Beat Saber) XR application contexts.

Main Results:

• Carefully paired multimodal privacy mechanisms reduced re-identification rates from over 80% to approximately 26% in casual and competitive XR applications.
• All evaluated paired mechanisms met established performance thresholds for real-time interaction.
• The XR Privacy SDK was released to aid developers in implementing these privacy solutions.

Conclusions:

• Multimodal privacy strategies are crucial for effectively protecting behavioral data in XR environments.
• Context-aware and modality-specific privacy approaches offer significant improvements over naive, unimodal solutions.
• The findings support the development and adoption of advanced privacy techniques for immersive technologies.