WiDIGR: Direction-Independent Gait Recognition System Using Commercial Wi-Fi Devices

• Published in: IEEE internet of things journal Vol. 7; no. 2; pp. 1178 - 1191
• Main Authors: Zhang, Lei, Wang, Cong, Ma, Maode, Zhang, Daqing
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.02.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Channel state information (CSI); Computer Science; device-free sensing; Feature extraction; Fresnel model; Fresnel reflection; Gait recognition; Indoor environments; Legged locomotion; Sensors; Signal processing; Signal quality; Spectrogram; Ubiquitous Computing; Walking; Wireless fidelity; Channel state information (CSI); Fresnel model; Device-free sensing; Gait recognition
• ISSN: 2327-4662; 2372-2541; 2327-4662
• DOI: 10.1109/JIOT.2019.2953488

Gait recognition enables many potential applications requiring identification. Wi-Fi-based gait recognition is predominant because of its noninvasive and ubiquitous advantages. However, since the gait information changes with the walking direction, the existing Wi-Fi-based gait recognition systems require the subject to walk along a predetermined path. This direction dependence restriction impedes Wi-Fi-based gait recognition from being widely used. In order to address this issue, a direction-independent gait recognition system, called WiDIGR is proposed. WiDIGR can recognize a subject through the gait no matter what straight-line walking path it is. This relaxes the strict constraint of the other Wi-Fi-based gait recognition. Specifically, based on the Fresnel model, a series of signal processing techniques are proposed to eliminate the differences among induced signals caused by walking in different directions and generate a high-quality direction-independent signal spectrogram. Furthermore, effective features are extracted both manually and automatically from the direction-independent spectrogram. The experimental results in a typical indoor environment demonstrate the superior performance of WiDIGR, with mean accuracy ranging from 78.28% for a group of six subjects to 92.83% for a group of three.