Non-Contact Heart Rate Estimation From Photoplethysmography Using EEMD and Convolution-Transformer Network

Remote heart rate measurement aims to use the remote photoplethysmography (rPPG) technology to extract the heart rate information from face videos in a non-contact manner. Most of the existing methods are susceptible to environmental factors and the movements of the subjects. To address this challenge, we propose a lightweight convolutional neural network for real-time heart rate detection under realistic conditions. Here, the rPPG signals are decomposed using Ensemble Empirical Mode Decomposition (EEMD) technology to extract the Intrinsic Mode Functions (IMFs) that contain valuable heart rate features. Then, feed the extracting IMFs into a parallel CNN-Transformer network, which consists of two branches: the CNN branch is used to extract local features of the rPPG signal, while the Transformer branch is employed to capture the global representation of the rPPG signal. And then, heart rates are calculated using power spectral density. Extensive experiments are conducted on two public datasets – the UBFC-rPPG dataset and PURE dataset. In terms of the experiment results, the proposed method achieves 0.87 MAE (bpm), 1.72 RMSE (bpm) and 0.99 r value of Pearson’s correlation coefficient on UBFC-rPPG dataset, and 0.81 MAE (bpm), 1.65 RMSE (bpm) and 0.99 r value of Pearson’s correlation coefficient on PURE dataset.