A privacy-preserving approach to streaming eye-tracking data.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Virtual Reality

Background:

• Eye-tracking technology is increasingly integrated into mixed reality (MR) devices.
• This integration enables critical applications but poses significant privacy risks.
• Unique user identification is possible even under natural viewing conditions in virtual reality (VR).

Purpose of the Study:

• To assess the risk of unique user identification via eye-tracking data in MR.
• To propose and evaluate a framework for mitigating privacy risks associated with eye-tracking data.
• To ensure privacy-by-design principles are applied to eye-tracking data flow in MR.

Main Methods:

• Developing a framework incorporating API gatekeeping and software-implemented privacy mechanisms.
• Evaluating the effectiveness of the proposed mechanisms in reducing user identification rates.
• Measuring the impact of privacy mechanisms on gaze prediction accuracy.

Main Results:

• Eye-tracking data in VR presents an appreciable risk of unique user identification.
• The proposed framework reduced identification rates from 85% to 30%.
• The privacy mechanisms introduced less than 1.5° error in gaze position for gaze prediction.

Conclusions:

• User identification via eye-tracking in MR is a significant privacy concern.
• The developed framework effectively mitigates privacy risks without compromising essential functionalities like gaze prediction.
• This approach establishes the first privacy-by-design solution for eye-tracking data in MR use cases.