Breatheasy

A research prototype born from a personal need. It runs in the phone’s browser, captures the baby-monitor feed via screen capture, and applies a small motion-analysis pipeline to detect the rhythm of an infant’s breathing. Not a medical device, not aspiring to be one — an experiment in how far a camera and a signal-processing pipeline can get.

The real work was the measurement problem, not the plumbing. False positives are easy to generate — a fan, a fluttering blanket, a passing shadow all read as “motion” — so the pipeline is mostly signal conditioning: isolating the periodic component at breathing-plausible frequencies, gating on amplitude, and running detection through a dual-threshold hysteresis state machine so an alert only fires on a sustained breathing-to-lost transition, not a flicker. How I knew it worked without ever recording a child: the DSP and state machine are driven by synthetic-signal tests — sine waves in the breathing band versus pure noise — and on-device tuning used a numbers-only telemetry loop (thresholds and confidences, never frames).