Non-contact rPPG-based human status assessment via a spatial–temporal attention feature fusion network with anti-aliasing

Abstract

Remote Photoplethysmography (rPPG) is a cost-effective and non-contact technology that enables real-time monitoring of physiological status by extracting vital information such as heart rate (HR). This capability enables the assessment of fatigue and stress, helping to prevent accidents by identifying risky conditions early. Continuous monitoring with rPPG reduces operational risks, contributing to safer industrial and medical environments. However, the performance of rPPG is challenged by complex backgrounds and facial motions in industrial environments, which complicates feature extraction. To address these challenges, this paper proposes a spatial–temporal attention feature fusion network with anti-aliasing (ST-ASENet) for human status assessment. The ST-ASENet encodes spatial–temporal facial signals from multiple regions of interest (ROI) and enhances feature extraction through the attention mechanism. The network integrates anti-aliasing by low-pass filtering during the downsampling process to improve the accuracy of rPPG signals in complex environments. It calculates HR, respiratory rate (RR), and heart rate variability (HRV) for status evaluation. Additionally, the Robotics Operator Factors Assessment (ROFA) dataset is introduced, featuring diverse individuals and environments to improve the robustness of ST-ASENet. Experimental results demonstrate that ST-ASENet outperforms state-of-the-art methods in HR estimation and shows effectiveness across various industrial scenarios. The proposed method fosters operational efficiency and a data-driven approach to human-centric safety, making rPPG invaluable in modern, health-focused workplaces.