Wavelet-based RPPg signal analysis: cwt for spectral peak localization and identification with pulse oximeter validation

Abstract

Remote photoplethysmography (rPPG) has become a promising non-contact technology for cardiovascular monitoring, but the accuracy of spectral peak detection remains unpredictable due to motion artifacts, noise, and camera signal quality. Traditional methods often fail to localize and identify heart rate peaks in the presence of such disturbances. The aim of the study is to develop a wavelet approach to improve the reliability of rPPG spectral peak analysis by using a continuous wavelet transform (CWT) for accurate frequency-time localization, followed by systematic peak identification and verification using a medical-grade pulse oximeter. The rPPG signals were acquired under controlled conditions, processed using CWT to improve spectral characteristics, and subjected to a peak detection algorithm optimized for heart rate estimation. Wavelet coherence was used to evaluate the agreement between the peaks obtained with rPPG and the reference pulse oximeter data. The experimental results demonstrated that CWT-based peak localization achieved an average absolute error of 2.1 BPM compared to the pulse oximeter, with a coherence of 0.53 under steady-state conditions. The proposed method demonstrated improved robustness to motion artifacts compared to conventional Fourier-based approaches, especially in low-light or low signal quality scenarios. The proposed wavelet transform structure improves the accuracy and reliability of rPPG spectral peak detection, bridging the gap between non-contact measurements and clinical pulse oximetry. This research extends the potential of rPPG for real-world applications, such as remote health monitoring and wearable devices.