Heart Rate Imputation Using Accelerometers for Wearable Devices

Heart rate (HR) estimation from photoplethysmography (PPG) signals is a vital component of many health monitoring systems. However, PPG signals are susceptible to motion artifacts, which can lead to inaccurate HR measurements. Moreover, PPG-based HR estimation often encounters catastrophic failures in the presence of excessive motion artifacts where reliable HR reading from PPG is not possible. In this study, we introduce a novel approach to reducing catastrophic failures by imputing missing HR using tri-axial accelerometer data in consumer-grade smart watches. Our proposed method employs a lightweight neural network model that leverages accelerometer data and past HR measurements. Our experimental results on two datasets demonstrate that our method outperforms existing baseline methods by incorporating accelerometer data during HR imputation. Furthermore, the results show that our approach is particularly effective in certain scenarios, such as when activity transitions occur. This study has the potential to improve the reliability of health monitoring systems, making them more suitable for real-world use and providing accurate HR measurements even in challenging conditions.