Google’s AI team unveiled a system that measures heart rate passively using a phone’s front‑facing camera. Named PHRM, the pipeline captures eight‑second facial videos after each face‑unlock, runs a lightweight temporal‑shift CNN, and outputs a confidence‑scored HR estimate. In lab tests it achieved a mean absolute percentage error under 10%, meeting industry standards across all skin tones. Every unlock becomes a silent vital‑sign snapshot.

The research leveraged a dataset of 350,000 clips from 700 participants, deliberately balanced by the Monk Skin Tone scale: at least 25 % light, 25 % medium, and 33 % dark skin. Validation on a free‑living cohort of 231 users recorded 231 clips per day, comparing PHRM’s HR to ECG ground truth and to Fitbit Charge 6 data. All recordings respected user privacy through on‑device encryption.

Aggregating hourly readings through Kalman filtering produced a daily resting heart rate estimate within 5‑bpm mean absolute error of the Fitbit reference, surpassing 15 published rPPG models. Correlations between higher PHRM‑derived RHR and elevated BMI or lower VO2max confirmed clinical relevance. The release aims to accelerate open research. The team released the full video dataset and a pre‑trained PHRM‑mini model for qualified researchers.