Real-time vitals monitoring AI, processing rPPG computer vision telemetry, pulse detection, and clinical health dashboards.
Identified and resolved inaccuracies introduced when shifting the measurement window from 60s to 45s with a 5-second precheck. Recalibrated the processing pipeline to restore consistent vital estimation across variable camera feeds.
Version-locked the production rPPG processor and generated a 74-sample ground-truth dataset spanning diverse age groups. Quantified bias, MAE, and offsets for HR (+9.0 offset, 22.2 MAE), RR (+3.7 offset), SpO2 (+4.4 offset), SBP (-10.9 offset), and DBP (-1.7 offset, 82.4% acceptance rate).
Architected a parallel POS + CHROM signal extraction engine with a lightweight quality evaluator selecting the higher-confidence signal dynamically without modifying downstream vitals pipelines.
Implemented adaptive spatial-temporal filtering to eliminate ambient lighting noise and head movement distortions in real-world non-clinical environments.
Integrating multi-frame PPG waveform analysis for automated stress score and respiration rate predictions.
Automated bi-directional sync of captured vital metrics directly into hospital EHR systems.