Respiratory Rate Estimation Using Dual-IMU Signals and Deep Learning: A Spectrogram-Based Framework Toward Feasible Wearable Deployment

Abstract

Continuous, non-intrusive respiratory rate (RR) monitoring is often hindered by motion artifacts and unstable sensor–skin contact. We propose a dual-IMU waistband design and train a network to estimate RR from the sensor signals. In a 20-participant study spanning sitting, standing, walking, and running, our method consistently outperforms prior IMU-based approaches, particularly during dynamic motion. The results demonstrate that the network effectively separates respiratory signals from motion interference, providing accurate and reliable RR estimation with a minimal, wearable-friendly sensor configuration.