Article: Non-contact pulse rate estimation from remote photoplethysmography using wavelet transform filter and convolutional neural network Journal: International Journal of Biomedical Engineering and Technology (IJBET) 2025 Vol.49 No.4 pp.294 - 312 Abstract: Pulse rate (PR) measurement traditionally relies on contact sensors such as photoplethysmography (PPG). However, these are unsuitable during pandemics and for long-term monitoring. Remote photoplethysmography (rPPG) offers contactless cardiovascular monitoring by detecting blood pulsation-induced skin colour changes. While conventional rPPG methods suffer from poor accuracy due to motion and lighting sensitivity, deep learning approaches, though more accurate, require extensive datasets and computational resources while lacking interpretability. This study presents an improved hybrid approach combining conventional preprocessing with deep learning. Video data undergoes traditional processing, band-pass filtering to enhance PR frequencies, and continuous wavelet transform to generate time-frequency images. These feed into a streamlined convolutional neural network (CNN) which is designed to iteratively extract features, resulting in a network that is not overly complex for hardware deployment. Results demonstrate superior performance with 1.9 bpm RMSE, surpassing previous studies through the synergy of traditional preprocessing and CNN-based estimation. This research advances non-contact RGB camera applications in vital sign monitoring. Inderscience Publishers - linking academia, business and industry through research

Title: Non-contact pulse rate estimation from remote photoplethysmography using wavelet transform filter and convolutional neural network

Authors: Hoang Thi Yen; Doan Van Sang; Van-Phuc Hoang; Guanghao Sun

Addresses: Le Quy Don Technical University, Hanoi, Vietnam ' Vietnam Naval Academy, Khanh Hoa, Vietnam ' Le Quy Don Technical University, Hanoi, Vietnam ' The University of Electro-Communications, Tokyo, Japan

Abstract: Pulse rate (PR) measurement traditionally relies on contact sensors such as photoplethysmography (PPG). However, these are unsuitable during pandemics and for long-term monitoring. Remote photoplethysmography (rPPG) offers contactless cardiovascular monitoring by detecting blood pulsation-induced skin colour changes. While conventional rPPG methods suffer from poor accuracy due to motion and lighting sensitivity, deep learning approaches, though more accurate, require extensive datasets and computational resources while lacking interpretability. This study presents an improved hybrid approach combining conventional preprocessing with deep learning. Video data undergoes traditional processing, band-pass filtering to enhance PR frequencies, and continuous wavelet transform to generate time-frequency images. These feed into a streamlined convolutional neural network (CNN) which is designed to iteratively extract features, resulting in a network that is not overly complex for hardware deployment. Results demonstrate superior performance with 1.9 bpm RMSE, surpassing previous studies through the synergy of traditional preprocessing and CNN-based estimation. This research advances non-contact RGB camera applications in vital sign monitoring.

Keywords: camera-based; pulse rate; remote photoplethysmography; rPPG; wavelet filter; convolutional neural network; CNN.

DOI: 10.1504/IJBET.2025.150726

International Journal of Biomedical Engineering and Technology, 2025 Vol.49 No.4, pp.294 - 312

Received: 09 Apr 2025
Accepted: 29 Jul 2025
Published online: 22 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Non-contact pulse rate estimation from remote photoplethysmography using wavelet transform filter and convolutional neural network

Keep up-to-date