FVBPSR-Mamba: Facial video-based physiological signal recovery using Mamba for remote heart rate measurement

Remote photoplethysmography (rPPG) is an emerging non-contact modality that estimates vital signs by extracting blood volume pulse (BVP) signals from facial video sequences. Currently, the primary challenges in this field lie in extracting high-quality rPPG signals from video segments characterized by significant spatiotemporal redundancy and in accurately capturing the periodic patterns of rPPG within a long-term context.In this study, an end-to-end high-quality rPPG signal recovery system, termed FVBPSR-Mamba, is proposed for remote heart rate measurement via video. In this framework, a Multi-scale Hierarchical Spatial Mamba (MHSM) module is designed to explore subtle rPPG signals from multi-scale spatiotemporal receptive fields while enhancing spatial perception and temporal context understanding. Furthermore, a frequency-domain noise reduction module is incorporated to strengthen the quasi-periodic patterns of rPPG and mitigate interference from irrelevant noise. Extensive benchmarking against both traditional methods and state-of-the-art (SOTA) deep learning models demonstrates that the proposed approach achieves superior accuracy and robust generalization in both intra-dataset and cross-dataset evaluations on the UBFC-rPPG and UBFC-Phys datasets. In conclusion, the FVBPSR-Mamba system can effectively recover rPPG signals from facial videos, facilitating the remote measurement of vital signs such as heart rate.