Augmented and Mixed Reality (AR/MR) systems offer uniquely immersive and collaborative experiences, fundamentally diverging from traditional mobile interactions. As these technologies become more pervasive, ensuring user privacy is paramount. This paper addresses gait privacy, a critical concern where an individual's walking pattern can inadvertently reveal sensitive personal information like age, ethnicity, or health conditions. We introduce GaitGuard, a novel, real-time system designed to safeguard gait privacy against video-based gait profiling threats within MR environments. GaitGuard operates on a multi-threaded framework, incorporating dedicated modules for efficient stream capture, body detection and tracking, and effective privacy protection. Our rigorous evaluation involved testing 248 distinct configurations, systematically varying regions of interest, privacy techniques, and operational parameters. This comprehensive analysis allowed us to thoroughly assess the trade-offs between privacy protection, video quality, and system performance. Furthermore, we propose an innovative adaptive method that intelligently processes only gait-critical frames, significantly enhancing visual quality without compromising privacy for real-time deployment. GaitGuard demonstrates substantial privacy protection, achieving up to a 68\% reduction in gait profiling accuracy and inducing a significant feature distribution shift (Jensen-Shannon Divergence of 0.63). Crucially, the system maintains a high performance of 29 frames per second (FPS), ensuring an acceptable user experience. User studies with 20 participants further validate our approach, indicating greater user comfort and acceptance of the privacy-preserving transformations. GaitGuard offers a practical and immediately deployable solution for robust gait privacy in MR, without sacrificing the immersive user experience.

翻译：暂无翻译