| Title |
Supervised TS-CAN-Based rPPG Signal Acquisition and Stress Analysis Before and After Using VR Devices |
| Authors |
유래현(Rae-Hyun Yu) ; 김경호(Kyung-Ho Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.6.1122 |
| Keywords |
Deep Learning; Supervised; Two-Stream Convolutional Attention Network; Remote Photoplethysmography; Virtual Reality; Heart Rate Variability |
| Abstract |
The use of VR devices is gradually increasing due to the emergence of various applications such as therapy, research, and entertainment in virtual environments. However, one of the most common side effects of increasing usage is mental health, such as dizziness due to eye fatigue caused by high immersion. Therefore, this paper analyzes the health status through HRV(Heart Rate Variability) analysis by obtaining bio-signals using a non-contact method before and after using VR(Virtual Reality) devices. To acquire the signals, we used the TS-CAN(Two-Stream Convolutional Attention Network) method, which is a supervised learning among deep learning models, and used the UBFC-rPPG dataset and our own dataset, which is a face image taken from a webcam with a resolution of 640x480 at 30 fps. The rPPG (Remote Photoplethysmography) data obtained before and after using the VR (Virtual Reality) device were analyzed using HRV (Heart Rate Variability) analysis. The results showed that VR device usage led to a significant increase in BPM (Heart Rate) and LF Power, as well as a decrease in RMSSD, NN50, and SDNN, indicating a shift in autonomic nervous system balance towards sympathetic dominance, consistent with stress response. These findings suggest that VR usage has a notable impact on physiological stress markers, with implications for health monitoring in VR environments. This study emphasizes the utility of rPPG in assessing physiological responses to stress induced by VR usage. The key findings demonstrate that HRV metrics, such as BPM and LF Power, significantly increased, while RMSSD and SDNN decreased, providing objective insights into the stress effects of VR environments. These results underline the potential of non-contact biosignal analysis for health monitoring in Virtual Reality applications. |