ScholarWorks@동국대학교

초록

This study aimed to develop a foundational technology for a real-time biosignal monitoring system applicable to the daily lives of older adults by achieving functional light-weighting through the minimization of sensor configurations. To this end, two deep learning-based regression models were implemented to estimate respiratory rate using EDR spectrograms derived from ECG signals, three HRV-based features (RMSSD, SDNN, and Mean RR interval), and ACC data. The experiment was conducted on 210 participants who performed 15 types of daily movements, and the collected biosignals were used to evaluate model performance. Accuracy was defined as the proportion of predictions whose absolute error from the ground truth respiratory rate fell within ±0.05 Hz, ±0.10 Hz, and ±0.20 Hz error margins. As a result, the model using EDR spectrograms and HRV features showed the highest accuracy for fall movements, while the model using EDR and ACC inputs showed relatively high accuracy for daily life movements. This study demonstrates the potential to estimate respiratory rate in dynamic environments without the need for dedicated respiratory sensors. Future work will focus on expanding input data types and enhancing model architectures to further improve prediction accuracy and practical applicability. © 2025 IEEE.

키워드