Augmented/Mixed Reality (AR/MR) technologies usher in a new era of immersive, collective experiences, differentiating them from traditional mobile systems. As these technologies evolve, prioritizing privacy and security is critical. This paper focuses on gait privacy, where gait, the way a person walks, can reveal sensitive information such as age, ethnicity, or disorders. We present GaitGuard, a real-time system that protects gait privacy against video-based gait extraction attacks in MR environments. GaitGuard leverages a multi-threaded framework to efficiently process video frames, incorporating dedicated modules for stream capture, body detection and tracking, and privacy leak mitigation. We compare and combine multiple mitigation techniques, offering guidance to navigate the privacy-utility tradeoff. Through extensive experiments covering 248 settings across mitigation regions, types, and tunable parameters, we assess the impact of these techniques on privacy, video quality, and system performance. GaitGuard reduces the confidence of video-based gait extraction attacks by introducing a substantial distribution shift (Jensen-Shannon Divergence of 0.63, indicating highly altered gait features) and a decrease in identification risks by up to 68%, while maintaining 29 FPS and preserving video clarity. GaitGuard provides a practical real-time solution for privacy-preserving MR applications without affecting the MR user experience based on 20 subjective user surveys.
View on arXiv@article{romero2025_2312.04470,
title={ GaitGuard: Towards Private Gait in Mixed Reality },
author={ Diana Romero and Ruchi Jagdish Patel and Athina Markopoulou and Salma Elmalaki },
journal={arXiv preprint arXiv:2312.04470},
year={ 2025 }
}