Abstract
Prior research has validated Photoplethysmography (PPG) as a promising biomarker for assessing stress factors in construction workers, including physical fatigue, mental stress, and heat stress. However, the reliability of PPG as a stress biomarker in construction workers is hindered by motion artifacts (MA) - distortions in blood volume pulse measurements caused by sensor movement. This paper develops a deep convolutional autoencoder-based framework, trained to detect and reduce MA in MA-contaminated PPG signals. The framework's performance is evaluated using PPG signals acquired from individuals engaged in specific construction tasks. The results demonstrate the framework has effectiveness in both detecting and reducing MA in PPG signals with a detection accuracy of 93% and improvement in signal-to-noise ratio by over 88%. This research contributes to a more reliable and error-reduced usage of PPG signals for health monitoring in the construction industry.
Original language | English (US) |
---|---|
Article number | 105537 |
Journal | Automation in Construction |
Volume | 165 |
DOIs | |
State | Published - Sep 2024 |
Keywords
- Anomaly detection
- Autoencoder
- Construction worker health monitoring
- Convolutional neural network
- Motion artifact
- Photoplethysmography
ASJC Scopus subject areas
- Control and Systems Engineering
- Civil and Structural Engineering
- Building and Construction