Heart Rate Sensing Using Adaptive Motion-Compensated Contactless rPPG From Facial Imaging

Cardiovascular health monitoring is essential for early disease detection and daily health management. Traditional contact-based methods face challenges such as discomfort. Remote photoplethysmography (rPPG), a contactless heart rate (HR) detection technology based on computer vision, is a promising alternative but struggles in dynamic environments with facial movements and varying lighting. Unlike conventional algorithms hindered by low signal-to-noise ratios and supervised learning methods limited by insufficient training-data diversity, this study presents an enhanced motion-robust rPPG sensing framework that integrates adaptive region-of-interest (ROI) weighting, pyramidal Lucas–Kanade (LK) optical flow-based motion compensation, and an enhanced orthogonal matrix image transformation (E_OMIT) signal-processing pipeline. With a 3-D face mesh for initial ROI calibration and pyramidal LK optical flow for continuous facial tracking, the framework improves robustness against facial movements and monitoring accuracy. Under dynamic conditions, the proposed method reduces the mean absolute error (MAE) by 45.6% on UBFC-Phys and achieves 4.89 bpm (33.7% lower) on electrocardiography-Fitness (ECG-Fitness) versus the original orthogonal matrix transformation (OMIT). The ablation study further validates motion compensation and filtering, yielding an MAE of 2.41 bpm and a Pearson correlation coefficient (PCC) of 0.93. These experimental results demonstrate that the integration of motion compensation, adaptive regional fusion, and signal postprocessing substantially improves the robustness of contactless HR estimation under facial motion. The application demonstration further confirms these findings, revealing significant advancements in noncontact physiological monitoring during facial movements.