Non-Contact Rppg-Based Human Status Assessment Via Feature Fusion Embedding Anti-Aliasing in Industry

Abstract

Real-time non-contact monitoring of human physiological status through remote Photoplethysmography (rPPG) is emerging as a noteworthy cost-effective technology in industry. Early monitoring of human status enables timely interventions to enhance safety by preventing accidents and health issues. However, the rPPG performance is often limited due to complex background and facial motions in industrial environments, which introduce distractions and variations in facial landmarks, complicating the extraction of facial features. To overcome these challenges, we propose a novel spatio-temporal attention feature fusion network embedded with anti-aliasing (ST-ASENet) for human status assessment in industrial settings. This network utilizes spatialtemporal representations encoding facial signals from multiple regions of interest (ROIs) as input, which are processed through an attention network to enhance feature extraction capabilities for HR estimation. Moreover, the integration of low-pass filtering during the downsampling phase mitigate aliasing effects, enabling more accurate facial information extraction in complex environments. Then, the RR and HRV are calculated based on the reconstructed high-precision rPPG signal providing a comprehensive evaluation of human status. Furthermore, we have established the dataset of robotics operator factors assessment (ROFA), which includes video and HR data from individuals of diverse genders and nationalities, captured in both typical and industrial environments, verifying the robustness of the network. Extensive experiments on datasets show that proposed ST-ASENet not only outperforms the state-of-the-art (SOTA) on HR estimation but also exhibits promising model robustness in crossdataset scenarios, confirm the superior performance of our approach in industrail settings.

Keywords: remote photoplethysmographyhuman statusattention mechanismanti-aliasingspatial-temporal representations