Vitals™ SDK Documentation

Appearance

Research Publications & Clinical Validation ​

Vital Signs ​

Contactless Blood Oxygen Saturation Estimation from Facial Videos Using Deep Learning ​

Year: 2024

Objective: To develop and evaluate deep learning models capable of remotely measuring blood oxygen saturation (SpO₂) levels using facial videos captured with standard RGB cameras.

Key Messages:

  • An end-to-end deep learning model was developed to extract SpO2 from facial videos using spatial-temporal maps.
  • It achieves a mean absolute error (MAE) of 1.274% and a root mean squared error (RMSE) of 1.71%, surpassing the international standard of 4% for approved pulse oximeters.

Reference: MDPI

Robust Heart Rate Variability Measurement from Facial Videos ​

Year: 2023

Objective: To develop a contactless method for accurately measuring heart rate variability (HRV) using remote photoplethysmography (rPPG) from facial videos.

Key Messages:

  • WaveHRV was introducted which is based on the Wavelet Scattering Transform, adaptive bandpass filtering, and inter-beat-interval analysis.
  • WaveHRV achieved a mean absolute error (MAE) of 10.5 ms for the root mean square of successive differences (RMSSD) and 6.15 ms for the standard deviation of normal-to-normal intervals (SDNN) on the UBFCrPPG dataset.

Reference: MDPI

Vitals: Camera-based Physiological Monitoring and Health Management Platform ​

Year: 2022

Objective: To provide a practical, cost-effective solution for remote measurement of vital signs and health management. The platform is designed for easy deployment on everyday devices, facilitating seamless integration into daily routines and promoting proactive health monitoring.

Key Messages:

  • "Vitals" demonstrates the potential of camera-based physiological monitoring as a tool for pervasive healthcare. It allows users to gain insights into health trends and risk factors.
  • Validated against FDA-approved contact-based devices, "Vitals" has achieved promising accuracy exceeding 98% for selected vital signs, underscoring its reliability as a non-invasive health monitoring solution.

Reference: IEEE Xplore

Camera-based heart rate variability and stress measurement from facial videos ​

Year: 2022

Objective: To extract heart rate variability (HRV) by analyzing subtle light variations on the skin using an RGB camera and analyze the stress states of the subjects.

Key Messages:

  • A novel method for HRV measurement from rPPG signals was developed.
  • The results demonstrated the potential for an off-the-shelf webcam to provide robust HRV measurement and subsequent stress estimation.

Reference: IEEE Xplore

Deep Learning Methods for Remote Heart Rate Measurement: A Review and Future Research Agenda ​

Year: 2022

Objective: To provide a comprehensive overview of recent advancements in deep learning (DL) techniques applied to remote heart rate (HR) measurement. It seeks to categorize these methods based on their model architectures and applications, discuss their real-world implementations, and identify existing research gaps to guide future explorations in the field.

Key Messages:

  • This review highlights the significant progress made in contactless HR monitoring through remote photoplethysmography (rPPG) and the integration of DL methods. It categorizes DL approaches into end-to-end and hybrid methods, detailing their respective architectures and functionalities.
  • The paper also emphasizes the potential of these technologies in various real-world applications, such as healthcare and fitness monitoring.

Reference: MDPI

Improving rPPG Extraction ​

An image enhancement based method for improving rPPG extraction under low-light illumination ​

Year: 2025

Objective: To develop an image enhancement technique that can preprocess facial videos captured under low-light conditions, thereby improving the accuracy and reliability of rPPG-based heart rate measurements.

Key Messages:

  • The research introduces a novel image enhancement framework that significantly improves the quality of facial videos taken in low-light settings, facilitating more effective rPPG signal extraction.
  • By applying this enhancement method, the study demonstrates a notable improvement in the signal-to-noise ratio of the extracted rPPG signals, leading to more accurate heart rate estimations.

Reference: ScienceDirect

DiffPhys: Enhancing Signal-to-Noise Ratio in Remote Photoplethysmography Signal Using a Diffusion Model Approach ​

Year: 2024

Objective: To improve the quality of remote photoplethysmography (rPPG) signals, which are used for non-contact cardiovascular monitoring through facial videos.

Key Messages:

  • DiffPhys, a novel deep generative model based on a conditional diffusion model was introducted to enhance the SNR of rPPG signals.
  • This enhancement facilitates more accurate monitoring of health conditions in non-clinical settings, showcasing the potential of diffusion model approaches in improving the robustness and reliability of rPPG-based physiological measurements.

Reference: MDPI

Deep learning-based image enhancement for robust remote photoplethysmography in various illumination scenarios ​

Year: 2023

Objective: To improve deep learning-based rPPG extraction under different illumination conditions

Key Message: A retinex theory-based image enhancement model was developed to isolate illumination map and help improving rPPG extraction under low-light illumination conditions.

Reference: Google Scholar

Optimising rPPG Signal Extraction by Exploiting Facial Surface Orientation ​

Year: 2022

Objective: To enhance the accuracy of remote photoplethysmography (rPPG) by investigating how the orientation of facial surfaces affects signal quality.

Key Messages:

  • The research demonstrates that facial surface orientation significantly influences the quality of extracted rPPG signals. Regions with smaller angles of reflection, such as the cheeks and forehead, provide stronger signals.
  • By applying a threshold to the angle map, the study showcases the potential for dynamic ROI selection, thereby improving the rPPG signal extraction process.

Reference: CVF

Blood Pressure ​

Remote Blood Pressure Estimation from Facial Videos using Transfer Learning: Leveraging PPG to rPPG Conversion ​

Year: 2025

Objective: To develop a non-contact method for estimating blood pressure (BP) by utilizing facial videos. The study focuses on transferring knowledge from contact-based photoplethysmography (PPG) signals to remote photoplethysmography (rPPG) signals extracted from facial videos, with the goal of enabling accurate BP estimation without the need for physical contact.

Key Messages:

  • The research demonstrates that deep learning models trained on contact-based PPG data can be effectively adapted to estimate BP from rPPG signals obtained through facial videos.
  • The study reports mean absolute errors (MAE) for systolic and diastolic BP estimation, highlighting the potential of this approach for non-invasive BP monitoring.

Blood Pressure Measurement: From Cuff-Based to Contactless Monitoring ​

Year: 2022

Objectives:

  • To trace the evolution of blood pressure (BP) measurement techniques, starting from traditional cuff-based methods to modern contactless approaches.
  • To elucidate the biophysical principles underlying BP measurement, discuss the development of contact-based photoplethysmography (PPG) techniques, and highlight recent innovations in remote photoplethysmography (rPPG) algorithms that facilitate contactless BP monitoring.

Key Messages:

  • The paper underscores the significant advancements in BP measurement, noting a shift from conventional cuff-based devices to non-invasive, contactless technologies.
  • It emphasizes that recent developments in rPPG algorithms have enabled the extraction of vital cardiovascular parameters from facial videos, paving the way for convenient and continuous BP monitoring without physical contact.

Reference: MDPI

Remote Temperature Screening ​

Remote Mass Facial Temperature Screening in Varying Ambient Temperatures and Distances ​

Year: 2023

Objective: To improve the accuracy of remote facial temperature estimation for a crowd of subjects under the circumstance that the distance between sensor and subjects is long.

Key Messages:

  • A novel compensation model to enhance the accuracy of remote mass facial temperature screening systems.
  • Implementing this model reduced measurement error by 23.5% and extended the effective detection range by up to 46%, achieving sensitivity and specificity rates exceeding 90%.

Reference: CVF

An infrared thermography model enabling remote body temperature screening up to 10 meters ​

Year: 2021

Objective:

  • To enhance the accuracy and efficiency of remote body temperature measurements using infrared thermography and investigates the impact of sensor-to-subject distance on temperature readings.

Key Messages:

  • The research demonstrates that sensor-to-subject distance significantly affects the accuracy of temperature readings in remote screening scenarios.
  • A compensation model that adjusts temperature measurements based on distance.
  • Experimental results indicate that the model effectively identifies individuals with elevated temperatures.

Reference: CVF

Copyright © 2021-present PanopticAl Ltd. All Rights Reserved.