PhysFlow: Frequency Decoupled with Dual-Field Rectified Flow for Remote Photoplethysmography

2026-06-22

AI Summary

This work addresses the challenge of robust remote photoplethysmography (rPPG) signal recovery under complex real-world disturbances such as illumination variations, facial expressions, and head movements, which often introduce strong coupling artifacts that degrade pulse signal estimation. To this end, we propose PhysFlow, a novel framework that introduces frequency decoupling into rPPG modeling for the first time. PhysFlow decomposes the physiological signal into trend and amplitude components, each modeled by a dual-conditional velocity field. By integrating rectified flow with ordinary differential equation (ODE) solvers, the method enables efficient and robust waveform reconstruction. Extensive experiments demonstrate that PhysFlow significantly outperforms state-of-the-art approaches across multiple benchmark datasets, achieving notably improved heart rate estimation accuracy and waveform fidelity, especially in challenging scenarios with severe interference.

Abstract

Remote Photoplethysmography (rPPG) enables contactless pulse estimation from facial videos, serving as a vital tool for health monitoring. However, current deep learning methods often struggle under complex disturbances, particularly varying illumination, facial expressions, and unconstrained head movements. In such scenarios, subtle physiological signals are easily dominated by external interference, making the recovered rPPG waveform unstable and unreliable. One important reason is that most existing methods directly model the rPPG signal in a unified manner, where different signal components are coupled during reconstruction. This makes it difficult to preserve weak pulse-related variations when strong disturbance-induced changes are present. To address this challenge, we propose PhysFlow, a frequency-decoupled dual-field rectified flow framework tailored for robust rPPG estimation. Specifically, the ground-truth rPPG signal is decomposed into trend and amplitude components, which are used as separate supervisory targets. Based on the extracted facial features, PhysFlow learns two component-specific conditional velocity fields to model the two components separately. This design reduces mutual interference between different components and improves the robustness of rPPG reconstruction under complex disturbances. Moreover, the rectified flow formulation enables efficient waveform reconstruction with only a few ordinary differential equation (ODE) integration steps. Extensive experiments on multiple benchmark datasets demonstrate that PhysFlow outperforms state-of-the-art methods in both heart-rate estimation and rPPG waveform reconstruction across diverse challenging scenarios.

Problem

• remote photoplethysmography
• signal decoupling
• physiological signal recovery
• complex disturbances
• rPPG waveform instability

Innovation

• frequency decoupling
• dual-field rectified flow
• remote photoplethysmography
• component-specific modeling
• robust signal reconstruction