Researchers Find Torso Signals Beat Facial Video for Breathing Monitoring

Researchers have found that camera-based respiratory monitoring can estimate breathing rates more accurately from the upper torso than from the face. Published in Sensors, their work investigated how anatomical region and color-channel selection influence respiratory rate estimation, finding that a simple, untrained chrominance signal outperformed the other evaluated models.

Serene young woman practicing mindfulness and deep breathing exercises, sitting in lotus position on her sofa
Study: Colour-Channel Selection for Video-Based Respiratory Rate Estimation: A Diagnostic Study with a Single-Shot Held-Out Evaluation. Image Credit: AYO Production/Shutterstock.com

These findings highlight the potential of contactless monitoring and offer new insights into the use of remote photoplethysmography (rPPG) to measure vital signs without physical contact.

Why Facial rPPG Signals Struggle to Track Breathing

Camera-based physiological monitoring has been used in medical settings as a non-invasive method to measure vital signs. By employing video cameras, these systems capture physiological signals from subtle changes in a person's appearance over time.

Such contactless physiological monitoring often relies on remote photoplethysmography (rPPG), an optical sensing technique that detects subtle skin-color changes caused by variations in cardiovascular blood volume pulses. Although scientists originally developed rPPG to measure heart rate and pulse variability, they later found that respiration also affects optical signals through changes in skin color, vascular volume, and rhythmic body movements.

Existing camera-based respiratory monitoring methods have primarily focused on facial regions because face-detection algorithms are effective. While chest and abdominal movements provide mechanical indicators of breathing, their contribution to camera-based respiratory measurements has received less attention.

Two Video Datasets Put Contactless Respiratory Monitoring to the Test

To establish a consistent basis for comparison, researchers developed a controlled evaluation framework covering anatomical regions of

interest, color projections, and model architectures. Rather than relying on large deep learning models, they compared compact dual-stream models that used deterministic, color-based signals and required no training.

The evaluation utilized two video datasets: COHFACE, which contained 160 video clips from 39 subjects with reference measurements from respiration-belt sensors, and VitalVideos, comprising high-frame-rate biomedical recordings.

Three color-channel projections were assessed: chrominance-based projection (CHROM), plane-orthogonal-to-skin (POS), and the green-red-blue difference method (2G-R-B). These signals were extracted from the face and an upper-torso bounding box defined relative to the face. Grayscale spatial patches were also tracked alongside the color signals to assess movement-related information.

The processing pipeline employed Python 3.10, NumPy, SciPy, OpenCV, and MediaPipe for facial landmark detection. Video recordings were divided into fixed windows, filtered using a Butterworth bandpass filter covering 0.10–0.50 Hz, and resampled to 4 Hz. Respiratory rate was estimated by identifying the dominant spectral peak through a windowed periodogram. The study implemented five-fold, subject-disjoint validation to ensure reliable comparisons, preventing the same subjects from appearing in both training and testing partitions.

Torso Chrominance Signal Cuts Breathing Rate Error to 1.267 Breaths/Min

The experimental outcomes showed that selecting the appropriate anatomical region significantly influenced respiratory monitoring accuracy, more than using complex deep learning models. An untrained channel extracted from the upper torso achieved a mean absolute error (MAE) of 1.267 breaths per minute and a correlation coefficient of +0.795.

Download a free copy of this page here!

In comparison, facial color channels produced errors ranging from 3.693 to 3.912 breaths per minute, performing worse than the baseline of about 2.644 breaths per minute. Within the torso region, only the chrominance projection reliably isolated the respiratory rate; other color representations failed to distinguish breathing patterns from background noise.

Researchers conducted tests to determine whether the torso signal resulted from global lighting changes or an arbitrary choice of measurement region. Five stationary background patches produced an error of 4.359 breaths per minute, indicating that fluctuations did not account for the measured signal. Moving the region upward to the neck increased the error to 2.580 breaths per minute, supporting the importance of selecting the torso region.

A separate mechanism test examined whether the chrominance signal merely reflected physical chest movement. Full-resolution sub-pixel row displacement measurements produced an error of 2.917 breaths per minute.

However, removing the in-band mechanical motion component from the chrominance signal did not reduce its accuracy; the resulting error was 1.156 breaths per minute. This suggests that the respiratory information in the chrominance signal cannot be explained solely by physical chest movement.

Simple Camera-Based Monitoring Shows Potential for Lightweight Systems

The findings have implications for camera-based monitoring in hospitals, telemedicine, and smart home environments.

Since the primary respiratory signal can be extracted from an untrained chrominance stream over the upper torso, monitoring systems may operate without intensive deep learning models.

This approach could support implementation on edge computing devices and low-cost optical sensor modules, including neonatal incubators and contactless bedside monitors. The torso's reliability as a respiratory signal source also offers opportunities for privacy-preserving monitoring, allowing developers to blur or mask facial features while retaining the torso region for physiological measurements.

Torso Signals Outperform Trained Models, but Clinical Validation Remains

These findings demonstrate that reliable camera-based respiratory monitoring depends on appropriate anatomical region and color-channel selection and comprehensive evaluation rather than complex model architectures. The study showed that a single, untrained torso chrominance channel outperformed the tested complex multi-stream networks, emphasizing the importance of selecting the appropriate anatomical region and color projection.

Future work should examine how these contactless sensing methods perform in active clinical environments, with particular attention to variations in skin tone, fabric texture, and movement. Further improvements to optical motion estimation methods may also help address challenges associated with movement.

Applying consistent benchmark protocols will be crucial for evaluating performance and establishing the reliability of camera-based physiological monitoring in clinical settings.

Journal Reference

Buyung, R. A. et al. (2026). Colour-Channel Selection for Video-Based Respiratory Rate Estimation: A Diagnostic Study with a Single-Shot Held-Out Evaluation. Sensors, 26(22). DOI: 10.3390/s26227107, https://www.mdpi.com/1424-8220/26/22/7107.

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.

Muhammad Osama

Written by

Muhammad Osama

Muhammad Osama is a full-time data analytics consultant and freelance technical writer based in Delhi, India. He specializes in transforming complex technical concepts into accessible content. He has a Bachelor of Technology in Mechanical Engineering with specialization in AI & Robotics from Galgotias University, India, and he has extensive experience in technical content writing, data science and analytics, and artificial intelligence.

Citations

Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Osama, Muhammad. (2026, November 17). Researchers Find Torso Signals Beat Facial Video for Breathing Monitoring. AZoSensors. Retrieved on November 17, 2026 from https://www.azosensors.com/news.aspx?newsID=17007.

  • MLA

    Osama, Muhammad. "Researchers Find Torso Signals Beat Facial Video for Breathing Monitoring". AZoSensors. 17 November 2026. <https://www.azosensors.com/news.aspx?newsID=17007>.

  • Chicago

    Osama, Muhammad. "Researchers Find Torso Signals Beat Facial Video for Breathing Monitoring". AZoSensors. https://www.azosensors.com/news.aspx?newsID=17007. (accessed November 17, 2026).

  • Harvard

    Osama, Muhammad. 2026. Researchers Find Torso Signals Beat Facial Video for Breathing Monitoring. AZoSensors, viewed 17 November 2026, https://www.azosensors.com/news.aspx?newsID=17007.

Tell Us What You Think

Do you have a review, update or anything you would like to add to this news story?

Leave your feedback
Your comment type
Submit

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.