GaitWay: Monitoring and Recognizing Gait Speed Through the Walls
Chenshu Wu, Feng Zhang, Yuqian Hu, K. J. Ray Liu
Abstract
Interests in monitoring and recognizing gait have surged significantly over the past decades. Traditional approaches rely on camera array, floor sensors (e.g., pressure mats), or wearables (e.g., accelerometers), none of which are suitable for continuous and ubiquitous everyday use. In this article, we present GaitWay, the first system that monitors and recognizes an individual's gait through the walls via wireless radios. GaitWay passively and unobtrusively monitors an individual's gait speed by a single pair of commodity WiFi transceivers, without requiring the user to wear any device or walk on a restricted walkway. On this basis, GaitWay automatically identifies stable walking periods, extracts physically plausible and environmentally irrelevant speed features, and accordingly recognizes a subject's gait. Built upon a distinct rich-scattering multipath model, GaitWay can capture one's gait speed when one is $>$ >10 meters away behind the walls. We conduct experiments in a typical indoor space and perform eight sessions of data collection with 11 subjects across six months, resulting in $>$ >5,000 gait instances. The results show that GaitWay achieves a median 0.12 m/s and 90%tile 0.35 m/s error in speed estimation, with a mean error of 3.36 cm in stride lengths. Further, it achieves a verification rate of 90.4% and a recognition rate of 81.2% for five users and 69.8% for 11 users, confirming its comfort and accuracy for continuous and ubiquitous use.