Deep learning-based robust heart rate estimation using remote photoplethysmography under different illuminations

Remote photoplethysmography (rPPG) monitors heart rate (HR) without requiring physical contact, which has applications. However, accurate measurement can be challenging due to different illuminations of the surrounding environment, and the impact of different illuminations on deep learning-based methods is more severe than that of traditional methods. In this study, to improve the robustness of the model to extract rPPG signals and estimate HR under different illuminations, we proposed a method called TNDR (temporal normalization and DC removal) to reduce lighting information in facial video and tested it using 3D CNN network (PhysNet). We evaluated our proposed method using four publicly available datasets (PURE, UBFC-rPPG, UBFC-PHYS, LGI-PPGI) and a proposed dataset containing three illuminations. The results show that the proposed method can significantly improve the accuracy of rPPG extraction and HR estimation under different illuminations.