→

How it works

How does rPPG work? How cameras measure vital signs

Vitals AI uses rPPG and rBCG to recover pulse-related changes in reflected light and tiny head movements from a short camera scan. This page follows the process from face detection to the 20+ markers a scan can produce.

On this page

The short answer

rPPG, or remote photoplethysmography, lets a camera recover a pulse signal from tiny changes in the light reflected by your skin. As blood volume changes with each heartbeat, the reflected signal changes too. Vitals AI also analyses rBCG, the small movements associated with each heartbeat, so one facial video can provide complementary optical and mechanical signals. The method needs a compatible camera and a clear view of your face, not a wearable or contact sensor.

What is rPPG?

PPG means photoplethysmography. It is an optical method for detecting changes in blood volume. In a pulse oximeter, a sensor sits on your fingertip or earlobe and measures changes in the light absorbed or reflected as blood moves through the vessels underneath. PPG has been used in clinical and consumer devices for decades.

rPPG means remote photoplethysmography. It applies the same principle without a sensor touching the skin. A camera reads tiny changes in the light reflected from your face, while software tracks those changes across a sequence of frames to recover a pulse signal. You may also see this called imaging PPG or iPPG in research literature.

A landmark 2008 study showed that an ordinary camera and ambient room light could recover a blood-volume pulse from facial video. Since then, research has expanded across heart rate, breathing rate, heart rate variability, motion and skin-tone robustness. The research and benchmarking evidence is covered on the Vitals AI research page.

A brief timeline:

1930s: photoplethysmography (PPG) emerges as an optical way to measure blood-volume changes.
2008: Verkruysse, Svaasand and Nelson recover a blood-volume pulse from facial video using an ordinary camera and ambient light.
2010s: research expands across heart rate, respiration, heart rate variability, motion and skin-tone robustness, and machine learning.
2020s: research expands into more varied, natural and clinically relevant recording settings.

rPPG is one part of how Upvio Vitals AI turns a short camera scan into a set of vital signs.

How does a camera measure heart rate?

A scan runs through a few software steps and does not need special hardware:

HOW A SCAN WORKS

Find the face

→

Track the signal

→

Clean it up

→

Derive the outputs

Find the face. The software locates your face and builds a 3D representation of its facial geometry. This creates a stable map of regions such as the forehead and cheeks, allowing the model to track those areas throughout the scan.
Track the signal. It follows frame-by-frame changes in reflected light within those regions. Each heartbeat produces a tiny, repeating shift in colour and brightness that is invisible to the eye but detectable across a sequence of frames. Those changes are combined into a pulse waveform.
Clean it up. Denoising and motion compensation help reduce interference from lighting shifts, small head movements and camera artefacts, isolating the underlying pulse signal.
Derive the outputs. The cleaned signal is used to calculate heart rate and other markers. Different outputs use different calculations, which are explained on what Vitals AI measures.

None of this needs a wearable, a finger clip or any sensor touching your skin. It needs a compatible camera and a clear, front-facing view of your face for the duration of the scan.

What does dual-stream measurement add?

Vitals AI also analyses rBCG, or remote ballistocardiography. It detects tiny head and body movements associated with each heartbeat. rBCG does not read blood flow directly. It provides a mechanical signal that complements the optical signal recovered through rPPG.

Together, the two streams give the model optical and mechanical information from the same video. This helps explain why Vitals AI uses both signals, while the accuracy of each output remains dependent on the metric and the recording conditions.

Rather than depending on one camera signal or one measurement pathway, Vitals AI evaluates multiple complementary signals in real time and applies quality controls to determine the most reliable result for each scan.

Illustrative animation of the optical and movement signals recovered from facial video by combining rPPG and rBCG.

What is the camera actually detecting?

At the core of rPPG is a simple physical change. Blood volume in the vessels near the skin changes with every heartbeat, and that change affects how much light the skin reflects. As blood volume rises and falls, the camera records a small repeating pattern in the video. Software separates that pattern from the rest of the image and uses it to recover a pulse signal.

The camera is reading ordinary reflected light. It does not need to touch the skin or use an invasive sensor.

Do I need a wearable or contact sensor?

No. That is the difference between rPPG and contact PPG. A pulse oximeter needs a sensor against your skin. rPPG needs only a compatible phone, laptop or webcam with a camera, plus a clear view of your face for a short scan. There is nothing to charge, wear or attach.

What conditions help a scan work well?

A clear scan depends on a few ordinary recording conditions:

Lighting. Even, adequate light helps the camera register the reflected-light changes clearly.
Movement. Staying reasonably still during the scan keeps the signal clean.
Camera quality. Resolution, frame rate, compression and exposure all affect how much physiological detail the camera can capture.
Face visibility. A clear, unobstructed view of the forehead and cheeks gives the software more region of interest to work with.

SETTING UP A CLEAR SCAN

Seated position

Even lighting

Camera at eye level

~30cm from camera

Face uncovered

Stay still

No talking

Rest 5 minutes first

On accuracy: Vitals AI uses an independently benchmarked rPPG and rBCG model evaluated against clinical reference devices. The full figures, methodology, population coverage and limitations are covered on Vitals AI research.

A few honest bounds apply across the board. Blood pressure from a camera scan is an estimate, not a cuff reading. Stress is a modelled score built from other measured signals, not measured directly. Vitals AI does not measure SpO2. Vitals AI is offered for general wellness use by default. A separate CE MDR Class IIa certified model is available for regulated applications.

Questions, answered

How does a camera measure heart rate?

It tracks tiny, heartbeat-driven changes in the light reflected from your face (rPPG), and in Vitals AI's case, small body movements from each heartbeat too (rBCG). Software finds your face, follows the signal across frames, filters out noise and motion, and converts the result into heart rate and other vitals.

Is rPPG the same as PPG?

They share the same optical principle: blood-volume changes alter reflected or absorbed light. PPG uses a sensor in contact with your skin, as in a pulse oximeter. rPPG (also called imaging PPG or iPPG) reads the same kind of signal remotely, using a camera instead of a contact sensor.

Do I need special hardware?

No. A compatible phone, laptop or webcam with a camera is enough. There is no wearable, finger clip or additional sensor to attach.

How long is a scan?

A scan takes about 30 seconds. During that time, the software analyses many video frames and several heartbeat cycles to recover the signal. Some outputs, such as heart rate variability, depend more heavily on the quality and duration of the captured signal.

Why do lighting and movement matter for a scan?

Because the whole method depends on reading a very small optical or movement signal. Good lighting and a reasonably still, visible face give the software a cleaner signal to work with.