Gait Authentication and Identification Using FMCW Radar: A Feature Extraction Framework

Read full text Summarise with VIRAT
Authors: Bruno Figueiredo, André Rouco, Beatriz Soares, Daniel Filipe Albuquerque, Ming Shen, Pedro Guilherme Lacerda Ferreira Pinho
Cited by: 0
References: 20

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Biometric authentication and identification based on gait offer a non-intrusive and effective solution for security and monitoring, as it can be performed remotely without physical contact or active user cooperation. Conventionalmethods such as fingerprint or facial recognition are more vulnerable to spoofing, while alternative user-intrinsic methods, such as electrocardio-gram (ECG), present disadvantages, since they require continuous physical contact with the user. In this context, frequency-modulated continuous-wave (FMCW) radars for gait analysis emerges as a promising alternative, enabling remote identification even under adverse visual conditions without compromising privacy. In this article, a one-dimensional (1D) FMCW radar system was developed to acquire a dataset comprising 98 participants across three days and three walking postures, in order to investigate the effects of intra and inter-subject variability. A signal-processing pipeline was implemented to extract and select gait-relevant features, revealing that only a subset remains stable across days. Two main scenarios were evaluated: authentication and identification. Authentication achieved 79.6% accuracy under cross-validation (CV), while identification reached 40%. Additional analysis, including conditiondependent evaluation and varying the number of users, demonstrated that performance is influenced by temporal variability and dataset size. In the Leave-One-Day-Out (LODO) scenario, authentication and identification achieved 70% and 13% accuracy, respectively, for the 98 subjects. When one walking posture was excluded, performance reached 76% for authentication and 26% for identification. Finally, the impact of the number of users on identification performance was evaluated, revealing a maximum accuracy of 98.9% for 2 participants, which decreased to 40% when all 98 were considered. These results highlight inter-day gait variability as the main limitation, with small changes, such as fatigue or footwear, significantly affecting micro-Doppler signatures and the resulting features.

Fields of Study & Subjects

Gait Recognition and Analysis Biometric Identification and Security Non-Invasive Vital Sign Monitoring

Publication details

Journal / source
IEEE Sensors Journal
Volume
26
Issue
9
Pages
13926–13939
Publication date
30 March 2026
Language
EN