Google's PHRM reads heart rate from your face via front camera
A Google Research system published in Nature estimates heart rate and resting heart rate from brief facial video captured passively during normal smartphone use.
Can your smartphone camera measure your heart rate without you doing anything?
Yes, according to new Google Research published in Nature. A system called PHRM captures short facial video clips via the front-facing camera right after face-unlock events, then applies deep learning to estimate heart rate. The approach requires no deliberate user action and no additional hardware beyond a standard smartphone.
How the measurement works
PHRM relies on photoplethysmography (PPG), the same optical technique used in pulse oximeters and wrist-worn fitness trackers. The front camera detects subtle changes in how light reflects off skin as blood pulses through it. The system processes 8-second facial video clips using a computationally efficient on-device pipeline, keeping the workload suitable for running in the background on consumer hardware.
Beyond single-point heart rate readings, PHRM aggregates measurements taken throughout the day to produce a daily resting heart rate (RHR) estimate. Resting heart rate is considered a meaningful biomarker for cardiovascular health; elevated RHR and upward trends over time have been linked to major adverse cardiovascular events and higher all-cause mortality.
Accuracy figures from the paper
| Metric | Result | Reference standard |
|---|---|---|
| Heart rate MAPE | < 10% | Electrocardiogram |
| Resting heart rate MAE | < 5 bpm | Wearable tracker |
The researchers report that the under-10% heart rate error meets industry accuracy standards and holds across people of all skin tones, a point the team specifically highlighted. Skin tone equity has been a persistent concern with PPG-based optical sensors, which can perform inconsistently on darker skin.
Why the front camera rather than a finger over the lens?
Google demonstrated finger-on-camera heart rate measurement in 2022, and later research explored whether that signal could help predict cardiovascular disease risk. The limitation of that approach is that it requires a deliberate gesture. PHRM targets what the team calls passive monitoring: measurements happen in the background, piggybacking on face-unlock events the user is already performing dozens of times a day.
The practical motivation is reach. Around five billion people already carry a smartphone, while wearable adoption remains lower, particularly in lower-resource settings and among populations at elevated cardiovascular risk. A system that works on the device already in someone's pocket could extend heart health tracking to people unlikely to buy a dedicated wearable.
Dataset and model released for researchers
Alongside the Nature paper, Google is releasing what it describes as the largest and most diverse publicly available dataset of smartphone facial videos for this kind of research. A pre-trained model called PHRM-mini accompanies the dataset. Both are gated: qualified researchers must apply for access rather than downloading freely, which is consistent with how sensitive biometric research datasets are typically handled.
The release follows a pattern Google Research has used before, publishing the methodology in a peer-reviewed journal while making artifacts available to the academic community to encourage follow-on work and independent validation.
What this is not
PHRM is described as a research system, not a shipping product feature. Google has not announced when or whether this capability will appear in Android, Pixel phones, or any consumer health application. The publication and dataset release position the work for external scrutiny and replication before any potential productization.