BIOWISH: Biometric Recognition using Wearable Inertial Sensors detecting Heart Activity
Executive Summary
This study introduces BIOWISH, a biometric recognition system leveraging wearable inertial sensors to detect mechanical heart activity via seismocardiography (SCG) and gyrocardiography (GCG). Using deep learning techniques such as WISHNETT and WISHNETT_F trained with siamese strategies, the system achieves high recognition performance across multi-session datasets, particularly when combining SCG and GCG data. The study also explores human activity recognition (HAR) as a precursor to biometric verification, achieving high accuracy through multi-modal approaches. Clinical implications include enhanced security in healthcare IoT frameworks and potential applications in continuous monitoring systems.
Answer Machine Insights
Q: What is the primary advantage of using SCG and GCG signals for biometric recognition?
SCG and GCG signals provide permanent characteristics that enable reliable biometric recognition even across multi-session datasets.
The permanence of discriminative characteristics within cardiac signals, a fundamental property for their usage in practical biometric recognition systems, has been addressed in several studies relying on medical-grade equipment.
Q: How does the system handle recognition across different activities?
The system employs a two-step process where human activity recognition (HAR) is performed first to identify the activity, followed by activity-specific biometric verification.
A two-stage process, requiring to determine the carried out activity before extracting person-specific characteristics, has been also proposed to perform automatic people recognition without having to know a priori the performed activity.
Key Results
Achieved equal error rates (EERs) below 5% for biometric recognition using SCG and GCG signals lasting 20 seconds.
Human activity recognition (HAR) accuracy exceeded 93% when combining SCG and GCG data.
Visual Evidence
![Fig. 1. Recordings of ECG, SCG (z-axis), GCG (y-axis), PCG, PPG, and BCG waveforms for a healthy subject. Relevant events include the mitral valve closure (MC) and opening (MO), the aortic valve closure (AC) and opening (AO), the S1 sound at the closure of the atrioventricular (mitral and tricuspid) valves, the S2 sound at the closure of the semilunar (aortic and pulmonary) valves, the systolic nadir (A) and end diastole (B). The H wave head-ward deflection begins close to the peak of the R wave and has maximum peak near the start of ejection; the I wave foot-ward deflection follows the H wave and occurs early in systole; the J wave largest head-ward wave follows the I wave and occurs late in systole; the K wave foot-ward wave follows the J wave and occurs before the end of systole (adapted from [23]).](/_next/image?url=%2Fscience-hub%2Fimages%2F10.1109_tdsc.2023.3268360%2F10.1109_tdsc.2023.3268360_p2_i1.png&w=3840&q=75)
Fig. 1. Recordings of ECG, SCG (z-axis), GCG (y-axis), PCG, PPG, and BCG waveforms for a healthy subject. Relevant events include the mitral valve closure (MC) and opening (MO), the aortic valve closure (AC) and opening (AO), the S1 sound at the closure of the atrioventricular (mitral and tricuspid) valves, the S2 sound at the closure of the semilunar (aortic and pulmonary) valves, the systolic nadir (A) and end diastole (B). The H wave head-ward deflection begins close to the peak of the R wave and has maximum peak near the start of ejection; the I wave foot-ward deflection follows the H wave and occurs early in systole; the J wave largest head-ward wave follows the I wave and occurs late in systole; the K wave foot-ward wave follows the J wave and occurs before the end of systole (adapted from [23]).
Clinical Snapshot
Evidence Rating
Relevance
high Priority