From promising demo to credible measurement infrastructure

A transparent validation roadmap often creates more confidence than a vague claim that the technology is accurate under all conditions. Here is what we test, how we test it, and what we report.

Four validation principles

We validate what we show

Every claimed capability has a defined validation study. We do not present results that have not been tested.

We report failure modes

Validation reports include rejection rates, edge case failures, and conditions where accuracy degrades.

We separate measured from calculated

Camera-measured values (HR, RR) are validated separately from calculated references (BMI, REE).

We do not claim under all conditions

Validation scope is explicit. Results outside tested conditions are not claimed.

Structured progression from synthetic to real-world

P1
Phase 1Active

Synthetic Signal Tests

Known-frequency synthetic signals injected into the pipeline. Validates that the rPPG algorithm extracts the correct frequency under ideal conditions.

Injected HR signals at 40–120 bpm
Injected RR signals at 8–30 /min
Mean absolute error measured
Edge cases: very low and very high HR
Outcome:Baseline algorithm correctness under ideal conditions
P2
Phase 2Active

Controlled Recording Tests

Pre-recorded videos with known ground truth, captured in stable lab conditions. Tests the pipeline against real camera data without confounding real-world variables.

Varied skin tones, ages, face sizes
Different lighting temperatures
Known HR from ECG reference
Multiple camera resolutions
Outcome:Performance across demographic and hardware variation
P3
Phase 3Upcoming

Reference Device Comparison

Simultaneous Camera Vital Check scan and validated reference device measurement. Direct comparison of HR and RR values.

Pulse oximeter or ECG for HR reference
Capnography or chest belt for RR reference
Bland-Altman analysis
Correlation coefficients reported
Outcome:Quantified agreement with validated reference devices
P4
Phase 4Upcoming

Real-World Usability Study

Real users, real hardware, real environments. Measures valid scan rate, rejected scan rate, and rescan patterns.

Valid scan rate
Rejected scan rate and reason distribution
Average time to valid scan
User satisfaction rating
Outcome:Real-world reliability and usability metrics
P5
Phase 5Upcoming

Fairness & Robustness Study

Performance across skin tones, lighting conditions, devices, and demographics. Identifies systematic bias or accuracy variation.

Fitzpatrick scale skin tone range
Varied lighting: warm, cool, bright, dim
Mobile, laptop, and desktop cameras
Age range: 18–65+
Outcome:Documented performance boundaries and fairness metrics

How we report results

Mean Absolute Error (MAE)
Primary accuracy metric for HR and RR
Root Mean Square Error (RMSE)
Sensitivity to outlier measurements
Bland-Altman Limits of Agreement
Agreement between CVC and reference device
Pearson / Spearman Correlation
Correlation between CVC and reference values
Valid Scan Rate
Percentage of scans returning a valid result
Rejected Scan Rate
Percentage and reason distribution of rejections
Moderate Quality Rate
Scans returned with quality note
Condition-stratified Results
Accuracy split by lighting, motion, skin tone