Multi-scene low-light remote physiological measurement database

• Published in: Machine vision and applications Vol. 36; no. 5; p. 100
• Main Authors: Wu, Junjun, He, Zuxian, Wang, Ying, Jiang, Yan, Cheng, Xu
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2025; Springer Nature B.V
• Subjects: Communications Engineering; Computer Science; Datasets; Deep learning; Design; Head movement; Heart rate; Illumination; Image Processing and Computer Vision; Infrared imaging; Light; Lighting; Measurement techniques; Networks; Neural networks; Pattern Recognition; Physiology; Video; Remote photoplethysmography; Low-light; Dataset; Face video
• ISSN: 0932-8092; 1432-1769
• DOI: 10.1007/s00138-025-01719-3

Remote photoplethysmography (rPPG) has garnered significant research attention for its ability to estimate heart rate from facial videos, achieving promising results under controlled lighting conditions. However, in real-world scenarios, the coupling of lighting variations and motion artifacts poses complex challenges that current methods often overlook. To address this gap, we introduce the Multi-Scene Low-Light rPPG Dataset (MSLR), which can simulate realistic conditions where lighting and motion jointly impact rPPG performance. Specifically, MSLR includes three distinct illumination levels, and each level encompasses stable, talking, and head movement scenarios to reflect various real-world situations. In addition, MSLR captures the synchronized infrared video of the subject under the minimum light setting, providing supplementary data to enhance robustness. Based on MSLR, we propose a novel Low-Light rPPG-Preserving Enhancement (LPPE) method, which compensates for brightness disparities between low-light and normal-light videos while preserving the integrity of rPPG signals. Our dataset has been rigorously validated using a variety of traditional handcrafted and deep learning methods, offering valuable insights into rPPG performance under low-light conditions. We believe this work provides a significant step forward in deploying rPPG in real-world applications and serves as a valuable resource for the research community. Our code and dataset are available at https://github.com/wwenmaositu/MSLR .