CN

SHEN GANGGANG SHEN

Professor      

  • Professional Title:Professor
  • Gender:Male
  • Status:Employed
  • Department:School of Software Engineering
  • Education Level:Postgraduate (Doctoral)

Paper Publications

Current position: 英文主页 > Scientific Research > Paper Publications

Toward Robust Person Identification Using BCG Signals: A Multi-stage Fingerprinting Approach

Release time:2025-01-17
Hits:
Indexed by:
Journal paper
Document Code:
2503915
Journal:
IEEE Transactions on Instrumentation and Measurement
Included Journals:
SCI
Volume:
74
Key Words:
Ballistocardiogram (BCG) signals fingerprinting nonorthogonal space projection person identification progressive metric learning
DOI number:
10.1109/TIM.2025.3527591
Date of Publication:
2025-01-09
Impact Factor:
5.6
Abstract:
Repeated personal identity verification is a tedious but crucial everyday task for people to access sensitive information and private assets. Being able to protect individuals’ rights while alleviating the burden of repetitive verification, unobtrusive person identification methods are attracting increasing attention. Since it contains information reflecting a person’s cardiovascular activities, the ballistocardiogram (BCG) signal has emerged as a noninvasive measure for biometric recognition. However, complex components and diverse waveforms of BCG signals pose challenges in effectively extracting identity information. To tackle these problems, we present MuSFId, a novel multistage fingerprinting-based identification approach for BCG signals. First, we introduce a nonorthogonal space projection algorithm to precisely extract heartbeat components from composite signals, generating a refined heartbeat waveform. Subsequently, we implement progressive metric learning to project these heartbeat segments into latent spaces, facilitating the separation of features unique to each individual. Furthermore, we devise a fingerprinting strategy to create identifiers for accurate individual matching. To validate the effectiveness of MuSFId, we conducted experiments using the Kansas dataset and private data collected with a low-cost PVDF sensor. The results demonstrate that MuSFId significantly outperforms existing methods, achieving a recognition accuracy of over 99%, underling its promising application potential.