Science
How a quiet sensor sees you breathing.
Alertero measures heart rate, breath rate, and fall events without a camera, a microphone, or anything you have to wear. Here is how that is possible — and what we deliberately chose not to build.
The signal
The sensor emits a low-power millimeter-wave radar pulse and listens for the reflection. When the pulse hits a person, a small amount of it bounces back. The shape of that returning signal changes as the chest wall moves — outward when the lungs fill, inward when they empty, and a much smaller, faster oscillation as the heart beats.
Two physiological signals share that one waveform. Breathing is a low-frequency, large-amplitude motion at roughly 0.2 to 0.4 Hz. The heartbeat is a high-frequency, much smaller-amplitude motion near 1 Hz. They overlap in space but separate cleanly in the frequency domain, which is how we recover both from a single sensor.
What we measure
- Heart rate
- Beats per minute, derived from the high-frequency component of chest-wall motion. Reported as a one-minute rolling average, with a confidence band that widens when the person is moving. Suitable for trend monitoring; not a substitute for a clinical ECG.
- Respiratory rate
- Breaths per minute, derived from the low-frequency component of chest-wall motion. Useful for catching apnea episodes and long pauses overnight, where the rate drops below a personal baseline.
- Fall events
- Detected by combining a sudden velocity transition with a sustained posture change to a horizontal position on the floor. Both conditions must hold for several seconds before an alert fires — sitting down quickly does not trigger one. We tune for a low false-positive rate over high recall, on the principle that an alert system that cries wolf gets ignored.
- Motion patterns
- A daily activity histogram, aggregated across the room. Useful for spotting prolonged stillness, restlessness, or drift from a personal routine — the kinds of changes that often precede a health event by days or weeks.
What we deliberately do not do
The sensor has no camera and no microphone. This is a hardware choice, not a software setting. The device cannot be reconfigured later to capture imagery or audio because the components required to do so are not present on the board.
Vital sign extraction runs on the ESP32 itself. The raw radar waveform never leaves the device. What is sent upstream is a small set of derived metrics — a heart rate, a breath rate, a fall flag — not a continuous signal a third party could re-interpret.
Alerts are encrypted in transit. The optional cloud history is encrypted at rest with keys the household controls. A sensor that has lost internet still detects falls and triggers local alerts; only the relay to a remote caregiver requires connectivity.
Why ESP32
The ESP32 is a low-power microcontroller with enough headroom to run the signal-processing pipeline on-device. Choosing it over a more powerful single-board computer was a privacy decision: a smaller compute envelope forces a discipline of doing less, and of sending only what is necessary off the device. There is no spare capacity for video, no spare capacity for audio, no spare capacity for the kind of always-on uploading that other home sensors rely on.
That same constraint is what lets the device run from a single wall outlet and stay quiet — no fan, no heat, no visible status lights at night. A monitor that wakes the person it is meant to be watching has missed the point.