LoneHeart — Camera Pulse Research

Camera access requires HTTPS or localhost. If you opened this HTML directly from file://, serve it from http://localhost on desktop or deploy it over HTTPS.

Try your camera.

Use soft front light. Center your face, then start a capture.

Duration Camera

Camera is off. Enable the preview when you are ready.

YOUR CAMERA STAYS LOCALA moment to sit still.

Enable your camera to preview the framing. Nothing is recorded until you start.

Camera is off

Camera not starting? Quick fixes

Open this page in Safari, Chrome, Edge, or Firefox directly, rather than an embedded social-media browser. Open the standalone lab if the embedded view is blank.
In the site controls beside the address, allow Camera and reload. On iPhone, check this website’s Camera permission in Safari. On Android, check both Chrome’s app permission and this site’s permission.
Close apps using the camera, such as video meetings. On Windows, also check Settings → Privacy & security → Camera → Let desktop apps access your camera.
If tracking cannot load, fixed-region mode still works as an experiment. Keep your face centered; there is no automatic face detection in that fallback. Synthetic lessons work without camera access.

Trackingoff

Lighting--

Motion--

Capture-- fps

Signalwaiting

INPUT / not started

Your result

Pulse extraction, color-method agreement, and consistency over time. A weak or conflicting signal produces no number.

Pulse estimate

beats/min

Waiting for measurement

Breathing rate

breaths/min

Waiting for measurement

Acquisition quality0%

Try a synthetic trace first, or opt in to a camera preview.

Extracted pulse · normalized

The last 30 seconds of extracted signal, or the available shorter span. This is not an ECG.

Where the signal concentrates

Run a signal to inspect its frequency content.

Method cross-checks · bpm

POS color: —
CHROM color: —
Pulse timing: —

Consistency across the capture · bpm

Early: —
Middle: —
Late: —

Candidate frequencies explain the analysis. They are not accepted results or measures of clinical accuracy.

What this teaches

POS combines normalized color channels; windowed CHROM and pulse timing provide cross-checks.
Peaks at the search boundaries are withheld. Harmonics need support at the fundamental, and the rate must remain consistent across the capture.
Rejected frames and long gaps can invalidate a seemingly convincing peak.
Quality and agreement are heuristics, not accuracy probabilities. The waveform is not an ECG.

Camera mode: use soft front lighting, center your face, and keep still. Choose 45–60 seconds for the exploratory breathing estimate. Lighting, motion, skin appearance, device processing, and frame timing can all affect results; performance across people and devices has not been established.

Camera access stops after capture, on Stop, or when this page is hidden. There is no recording upload or persistent signal storage.

Open Full LoneHeart ↗

See the signal. Test the result.

Give the lab a known pulse, then inspect what it recovers. These simulated color traces use the same analysis as camera input. A result is calculated only when the signal and its cross-checks agree.

Choose a signal. Results appear below—no camera needed.

Challenge the signal

Known pulseSignal condition

Research demonstration only. Not clinically validated or FDA-authorized. Do not use it for diagnosis, screening, medication, or care decisions. Synthetic accuracy does not demonstrate accuracy on people.