A Decision-Based Motion Artifact Removal Framework for Robust Heart Rate Estimation from Nose PPG and its Embedded Implementation

This work presents a robust, motion-aware framework for heart rate estimation via smart eyewear. We compared three photoplethysmography signal wavelengths (Green, Red, and Infrared) acquired at the nose across static and dynamic conditions. Our results show that while Infrared offers superior stability in static conditions, it exhibits the most significant morphological degradation during motion, showing higher coupling with motion-related inertial signals compared to Green and Red wavelengths. Leveraging these findings, we designed a selective pipeline that triggers spectral subtraction only when specific motion thresholds are exceeded. Offline validation demonstrated a maximum mean absolute error reduction of approximately 8 bpm during high-intensity motion. The embedded implementation (10 ms latency) confirmed real-time feasibility in a single-subject evaluation, reducing estimation errors by tens of bpm during motion without compromising static signal integrity.