Gait identification using a new time-warped similarity metric based on smartphone inertial signals
- PDF / 1,154,291 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 53 Downloads / 222 Views
ORIGINAL RESEARCH
Gait identification using a new time‑warped similarity metric based on smartphone inertial signals Sougata Deb1 · Youheng Ou Yang2 · Matthew Chin Heng Chua1 · Jing Tian1 Received: 27 March 2019 / Accepted: 19 December 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract The evolution of smart devices and ubiquitous computing have paved the way for more intelligent user verification options. Gait pattern recognition has gained prominence due to their noninvasive and seamless verification capabilities without requiring dedicated attention from the user. This paper proposes a new gait identification approach by proposing a new time-warped similarity metric for memory-based gait pattern analysis. The proposed metric is incorporated into a computationally-efficient template matching framework for gait identification. The proposed approach outperforms the conventional approaches by achieving a user recognition error rate of 7.7% and equal error rate of 11.2% based on samples from 50 subjects in two benchmark datasets. Keywords Gait identification · Dynamic time warping · Person identification · Smartphone inertial signal
1 Introduction Authentication methods are a vital element of sensitive data management for a wide range of applications including personal smartphones and wearables (Ahmad et al. 2016). As usage of these applications across domains such as banking, airline and telecom becomes ubiquitous, the security of sensitive information for the end user becomes paramount (Jain and Ross 2007). Traditional password driven authentication methods are gradually ceding their way to more spontaneous and intelligent biometric methods such as physiological or behavioral authentication (Amin et al. 2014). Physiological biometric authentication methods such as retina scan or finger printing are well-established (Sanderson and Erbetta * Jing Tian [email protected] Sougata Deb [email protected] Youheng Ou Yang [email protected] Matthew Chin Heng Chua [email protected] 1
Institute of Systems Science, National University of Singapore, Singapore 119615, Singapore
Department of Orthopaedic Surgery, Singapore General Hospital, Singapore 169608, Singapore
2
2000). Recent developments such as face recognition (Li et al. 2019) or action recognition (Satyamurthi et al. 2018) have begun to gain prominence (Tan et al. 2006; Graves et al. 2013) and are in the process of being commercialized as industrial solutions. While the physiological authentication methods generally enjoy a high accuracy, these approaches are explicit, requiring the user to perform some deliberate or stereotyped task for authentication (Crawford et al. 2013) and depend heavily on ambient conditions (Damaševičius et al. 2016). This is where implicit authentication such as keystroke or gait-pattern based methods which occur without conscious efforts of the user can deliver a significant benefit in user experience (Deng et al. 2018; El-Alfy and Binsaadoon 2019; Lamiche et al. 2019; Tao et al. 2
Data Loading...
