EN ·
🌏 中文

Advances in Remote Photoplethysmography (rPPG): A 2025 Research Overview

🔌 Connecting to arXiv database interface for retrieval…

⏳ Retrieval returned partial information; please wait…

Ching-Yi Lai et al.’s UMCL: Unimodal-generated Multimodal Contrastive Learning for Cross-compression-rate Deepfake Detection introduces a novel framework for Cross-Compression-Rate (CCR) deepfake detection. By converting unimodal visual data into three complementary features—rPPG signals, temporal dynamics, and semantic embeddings from vision-language models—the framework uses Affinity-driven Semantic Alignment (ASA) to align features. A Cross-Quality Similarity Learning (CQSL) strategy further enhances robustness across varying compression levels. Published in IJCV (Sept 2024).

Real-Time Mobile Video Analytics for Pre-arrival Emergency Medical Services presents the TeleEMS system, which integrates audio and video data to support multi-party video streaming. It incorporates three real-time modules: domain-specific LLM-based symptom analysis (EMSLlama), rPPG-based vital sign monitoring, and a multimodal model (PreNet) for joint text-vital analysis.

Ba-Thinh Nguyen et al.’s Reperio-rPPG: Relational Temporal Graph Neural Networks for Periodicity Learning in Remote Physiological Measurement proposes a framework using relational convolutional networks and graph transformers to capture the periodic structure of physiological signals, complemented by CutMix data augmentation for improved generalization.

Taixi Chen et al.’s TYrPPG: Uncomplicated and Enhanced Learning Capability rPPG for Remote Heart Rate Estimation introduces an efficient algorithm based on Mambaout. Key innovations include Gated Video Understanding Blocks (GVB) and a Comprehensive Supervision Loss (CSL). Presented at IEEE WI-IAT 2025.

Xulin Ma et al.’s Non-Contact Health Monitoring During Daily Personal Care Routines releases the LADH dataset, containing synchronized RGB/IR video and physiological signals from 21 participants. The study demonstrates that fusing RGB and IR inputs improves heart rate estimation accuracy (MAE=4.99 BPM). Published in IEEE BSN 2025.

Fangling Jiang et al.’s Learning Knowledge-based Prompts for Robust 3D Mask Presentation Attack Detection proposes a knowledge-graph-based prompt learning framework, utilizing causal graph theory to eliminate spurious correlations. Published in TPAMI (Oct 2025).

Zahra Maleki et al.’s SkinMap: Weighted Full-Body Skin Segmentation for Robust Remote Photoplethysmography introduces a skin segmentation technique that uses weighted full-body detection to exclude artifacts like mouths and eyes, maintaining accuracy under motion and speech conditions.

Tianwen Zhou et al.’s Editing Physiological Signals in Videos Using Latent Representations proposes a framework for editing heart rate signals in videos using 3D VAE encoding and target heart rate prompts, enabling precise modulation (10 bpm MAE) while maintaining visual quality.

Shuyang Chu et al.’s To Remember, To Adapt, To Preempt: A Stable Continual Test-Time Adaptation Framework for Remote Physiological Measurement in Dynamic Domain Shifts introduces PhysRAP, a continual test-time adaptation framework that prevents catastrophic forgetting while avoiding over-adaptation via preemptive gradient modification.

Bo Zhao et al.’s PHASE-Net: Physics-Grounded Harmonic Attention System for Efficient Remote Photoplethysmography Measurement derives a theoretical model from Navier-Stokes equations, designing zero-computation axial exchange modules and gated TCNs to achieve state-of-the-art performance in lightweight models.

Constantino Álvarez Casado et al.’s Design, Implementation and Evaluation of a Real-Time Remote Photoplethysmography (rPPG) Acquisition System for Non-Invasive Vital Sign Monitoring presents an optimized system for low-power devices, capable of 30 FPS operation with a multi-threaded architecture supporting HTTP streams and RESTful APIs.

Konstantin Egorov et al.’s Gaze into the Heart: A Multi-View Video Dataset for rPPG and Health Biomarkers Estimation provides a large-scale dataset of 3,600 synchronized videos from 600 subjects. Presented at ACMMM 2025.

Jiankai Tang et al.’s Contact Sensors to Remote Cameras: Quantifying Cardiorespiratory Coupling in High-Altitude Exercise Recovery investigates cardiorespiratory coupling (CRC) in high-altitude recovery, validating rPPG-based measurements with a high correlation (Pearson r=0.96) to oximetry. Published in UbiComp 2025.

Joaquim Comas et al.’s BeatFormer: Efficient motion-robust remote heart rate estimation through unsupervised spectral zoomed attention filters introduces a lightweight model combining zoomed orthogonal complex attention with unsupervised spectral contrastive learning (SCL).

Kang Cen et al.’s Robust and Generalizable Heart Rate Estimation via Deep Learning for Remote Photoplethysmography in Complex Scenarios designs an end-to-end network using differential frame fusion and Temporal Shift Modules (TSM), achieving 7.58 MAE on MMPD.

Zhipeng Li et al.’s Exploring Remote Physiological Signal Measurement under Dynamic Lighting Conditions at Night: Dataset, Experiment, and Analysis releases the DLCN dataset, featuring 13 hours of video across 4 typical night scenarios.

Jiho Choi et al.’s Periodic-MAE: Periodic Video Masked Autoencoder for rPPG Estimation introduces a self-supervised framework that uses frame-mask sampling and physiological frequency constraints to learn high-dimensional spatiotemporal representations.

Jiyao Wang et al.’s Align the GAP: Prior-based Unified Multi-Task Remote Physiological Measurement Framework For Domain Generalization and Personalization presents the GAP framework, simultaneously addressing Multi-Source Domain Generalization (MSSDG) and Test-Time Personalization Adaptation (TTPA).

Jiyao Wang et al.’s Efficient Mixture-of-Expert for Video-based Driver State and Physiological Multi-task Estimation in Conditional Autonomous Driving develops VDMoE, a system using RGB and rPPG to monitor driver cognitive load and fatigue via a Mixture-of-Experts (MoE) architecture.

Jitesh Joshi et al.’s Efficient and Robust Multidimensional Attention in Remote Physiological Sensing through Target Signal Constrained Factorization introduces the Target Signal Constrained Factorization Module (TSFM) for synchronous estimation of rPPG and respiratory signals from RGB and thermal videos.

Yiping Xie et al.’s PhysLLM: Harnessing Large Language Models for Cross-Modal Remote Physiological Sensing introduces PhysLLM, using Text Prototype Guidance (TPG) for cross-modal alignment, achieving SOTA performance across four benchmarks.

Kegang Wang et al.’s Memory-efficient Low-latency Remote Photoplethysmography through Temporal-Spatial State Space Duality proposes ME-rPPG, requiring only 3.6 MB of memory and 9.46 ms latency for real-time inference.

Rufei Ma et al.’s RF-BayesPhysNet: A Bayesian rPPG Uncertainty Estimation Method for Complex Scenarios integrates Bayesian neural networks with variational inference to model aleatoric and epistemic uncertainty, enhancing reliability.

Banafsheh Adami et al.’s rPPG-SysDiaGAN: Systolic-Diastolic Feature Localization in rPPG Using Generative Adversarial Network with Multi-Domain Discriminator uses a multi-discriminator GAN to reconstruct complete systolic-diastolic PPG waveforms.

Gyutae Hwang et al.’s Phase-shifted remote photoplethysmography for estimating heart rate and blood pressure from facial video uses a two-stage framework (DRP-Net and BBP-Net) to estimate heart rate and blood pressure from phase-shifted signals.

Hang Shao et al.’s Remote Photoplethysmography in Real-World and Extreme Lighting Scenarios designs a video Transformer utilizing global interference sharing to enable outdoor rPPG monitoring.

Theodore Curran et al.’s Estimating Blood Pressure with a Camera: An Exploratory Study of Ambulatory Patients with Cardiovascular Disease validates rPPG for blood pressure monitoring in high-risk cardiovascular patients.

Bochao Zou et al.’s RhythmMamba: Fast, Lightweight, and Accurate Remote Physiological Measurement leverages State Space Models (Mamba) to achieve linear computational complexity, increasing throughput by 319%.

Bochao Zou et al.’s RhythmFormer: Extracting Patterned rPPG Signals based on Periodic Sparse Attention introduces periodic sparse attention to optimize long-sequence processing.

Zheng Wu et al.’s CardiacMamba: A Multimodal RGB-RF Fusion Framework with State Space Models for Remote Physiological Measurement fuses RGB and Radio Frequency (RF) signals to improve fairness and robustness.

Zijie Yue et al.’s Bootstrapping Vision-language Models for Self-supervised Remote Physiological Measurement uses frequency-guided vision-text generation for self-supervised learning (IJCV 2024).

Bingjie Wu et al.’s Semi-rPPG: Semi-Supervised Remote Physiological Measurement with Curriculum Pseudo-Labeling combines curriculum pseudo-labeling with consistency regularization (IEEE TIM 2024).

Alexey Protopopov et al.’s A Robust Remote Photoplethysmography Method achieves 1.95 BPM MAE using infrared cameras by combining multiple mathematical transforms.

Yuting Zhang et al.’s Advancing Generalizable Remote Physiological Measurement through the Integration of Explicit and Implicit Prior Knowledge bridges domain gaps by decoupling physiological features from noise.

Key Research Areas

  1. Multimodal Fusion & Generalization: Fusing RGB/IR/RF signals to enhance domain robustness.
  2. Lightweight Modeling: Optimizing efficiency for mobile/edge deployment.
  3. Dynamic Environment Robustness: Mitigating motion artifacts and lighting interference.
  4. Self-supervised Learning: Leveraging MAEs and VLMs to reduce annotation dependency.
  5. Multi-task Estimation: Simultaneous assessment of heart rate, blood pressure, and respiration.
  6. Clinical Validation: Real-world evaluation in healthcare settings.
  1. Multimodality: Shifting from pure RGB to sensor-fused (RGB-IR-RF) approaches.
  2. Learning Paradigms: Moving from fully supervised to self-supervised/unsupervised frameworks.
  3. Real-world Deployment: Transitioning from laboratory settings to dynamic, outdoor, and long-term monitoring.
  4. Holistic Assessment: Expanding beyond heart rate to comprehensive vital sign monitoring.
  5. Architectural Innovation: Adopting State Space Models (SSM) and efficient Transformers for real-time performance.
  6. Clinical Reliability: Increasing focus on domain generalization and validation for medical diagnostics.