Moonwalk presents a unique method for user recognition using the built-in accelerometer in headphones. The method is based on gait recognition, allowing users to establish their identity by walking for a short period. The approach uses self-supervised metric learning to train a model that provides a distinctive representation of a user’s 3D acceleration, without requiring retraining. The method was tested on 50 participants and achieved an average F1-score of 92.9% and an equal error rate of 2.3%. The study also evaluated the performance under various conditions, like different shoe types and surfaces. The paper proposes new directions for advancing passive authentication for wearable devices.
Publication date: 13 Feb 2024
Project Page: https://arxiv.org/abs/2402.08451v1
Paper: https://arxiv.org/pdf/2402.08451
