Caregiver Activity Recognition Leveraging Wearable Sensing Technologies

Japan is facing a critical issue of rapid population aging, which has increased the demand for caregiving services while the available workforce continues to decline. Caregivers are frequently burdened with excessive workloads, particularly due to the manual recording of daily assistance tasks. In this study, we propose a lightweight system that automatically recognizes caregiving activities using wearable 9-axis sensors. The system is implemented with M5Stack devices, Wi-Fi data transmission, MongoDB cloud storage, and Long Short-Term Memory (LSTM) networks for activity recognition. We evaluate the system through experiments covering 11 caregiving activities and achieve a recognition accuracy of up to 91%. These results demonstrate that wearable sensor-based recognition can be feasibly introduced into caregiving environments to reduce documentation burdens and improve work efficiency.