Passive Heart Rate Monitoring via Smartphone Camera¶
Source: Google Research Blog Date Published: June 4, 2026
Overview¶
Google's Passive Heart Rate Monitoring (PHRM) system leverages the front-facing camera already in use during face unlock to measure heart rate via remote photoplethysmography (rPPG) — the technique of detecting blood volume pulse changes from subtle facial color variations.
Training Data¶
The model was trained on an extensive dataset of >350,000 video clips collected from approximately 700 participants, making it one of the largest rPPG training efforts to date.
Performance Results¶
| Metric | Performance |
|---|---|
| MAPE vs ECG | <10% (meets ANSI/CTA-20651 standard) |
| RHR MAE vs Fitbit | <4.39 bpm |
| Group 1 (light) MAPE | 5.04% |
| Group 2 (medium) MAPE | 5.12% |
| Group 3 (dark) MAPE | 7.84% |
Inclusivity & Skin Tone Performance¶
Critical to the system's validity, Google evaluated performance across skin tones using the Monk Skin Tone Scale. PHRM achieved non-inferiority across all skin tone groups — the only model among 15 leading rPPG models to meet accuracy targets across all tones. Notably, Group 3 (darkest skin tones) showed a MAPE of 7.84%, which still meets the ANSI/CTA standard but reveals a remaining performance gap.
Additional Contributions¶
- Largest public rPPG dataset released to the research community
- PHRM-mini — a lightweight model for on-device inference
- Designed for passive, frictionless health monitoring (no active measurement required)
Limitations¶
- Motion artifacts during talking or activity degrade accuracy
- Lower success rate for darker skin tones, though non-inferiority was achieved
- Performance in real-world uncontrolled environments needs further study