Open Access

ARTICLE

GestureID: Gesture-Based User Authentication on Smart Devices Using Acoustic Sensing

by Jizhao Liu^1,2, Jiang Hui^1,2,*, Zhaofa Wang^1,2

1 School of Computer Science, Zhongyuan University of Technology, Zhengzhou, 450007, China
2 Henan Development and Innovation Laboratory of Industrial Internet Security Big Data (Zhongyuan University of Technology), Zhengzhou, 450007, China

* Corresponding Author: Jiang Hui. Email:

Sound & Vibration 2024, 58, 151-169. https://doi.org/10.32604/sv.2024.045193

Received 20 August 2023; Accepted 26 December 2023; Issue published 19 March 2024

Abstract

User authentication on smart devices is crucial to protecting user privacy and device security. Due to the development of emerging attacks, existing physiological feature-based authentication methods, such as fingerprint, iris, and face recognition are vulnerable to forgery and attacks. In this paper, GestureID, a system that utilizes acoustic sensing technology to distinguish hand features among users, is proposed. It involves using a speaker to send acoustic signals and a microphone to receive the echoes affected by the reflection of the hand movements of the users. To ensure system accuracy and effectively distinguish users’ gestures, a second-order differential-based phase extraction method is proposed. This method calculates the gradient of received signals to separate the effects of the user’s hand movements on the transmitted signal from the background noise. Then, the second-order differential phase and phase-dependent acceleration information are used as inputs to a Convolutional Neural Networks-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model to model hand motion features. To decrease the time it takes to collect data for new user registration, a transfer learning method is used. This involves creating a user authentication model by utilizing a pre-trained gesture recognition model. As a result, accurate user authentication can be achieved without requiring extensive amounts of training data. Experiments demonstrate that GestureID can achieve 97.8% gesture recognition accuracy and 96.3% user authentication accuracy.

---

Keywords

---