Estimation of Heart Rate Based on Remote Photoplethysmography (rPPG) from an RGB Camera Using The Plan-to-Orthogonal Skin (POS) Method, Bandpass Filter, and The Short-Time Fourier Transform

Abstract

The rPPG signal uses an RGB camera to acquire a non-contact PPG signal. It has potential for telemedicine and remote monitoring, as it doesn't require additional devices, making it more efficient and adaptable. However, movement and lighting variations remain challenges. Numerous studies have explored various approaches based on deep learning, image enhancement, and hybrid methods. Most have focused on developing complex model architectures or improving signal quality. Few have systematically examined all the heart rate estimation process stages. This study examined the heart rate estimation process from rPPG signals using several configurations. These ranged from combinations of classical algorithms to the incorporation of temporal modeling. The results show that the most optimal heart rate estimation was obtained for the entire face ROI using a combination of the POS algorithm and a bandpass filter. This had an MAE of 2.00 BPM and an RMSE of 3.01 BPM. Heart rate estimation in the frequency domain yielded more optimal results compared to the time domain.