Enhancing Heart Rate Monitoring Using Remote Photoplethysmography Based on Neural Network

This study explores non-contact vital signs estimation using remote photoplethysmography (rPPG) for heart rate monitoring. Unlike traditional methods requiring direct skin contact, rPPG captures blood volume changes via facial color variations, enabling unobtrusive monitoring with standard cameras. We propose a Neural Network (NN) model incorporating convolutional layers to enhance rPPG signal quality for accurate heart rate estimation. Experimentation, conducted on a diverse dataset, evaluates various architectures and batch sizes. Results show NasNetMobile as the top-performing architecture, particularly with a batch size of 1. Challenges arise with a batch size of 2, highlighting the need for further refinement. Future work aims to incorporate pre-trained models and increase batch size post-dataset publication.