A Low-Power Intelligent Wearable System with Multi-Sensors and Lightweight Machine Learning Algorithm for Motion-Status Monitoring

Health monitoring enabled by wearable device aids in the early warning of potential health issues, but wearable devices still face technical bottlenecks in multiple physiology acquisition, low-power consumption, intelligent data processing. To address these challenges in a holistic manner, this paper proposes a multi-sensor, intelligent and low-power wearable system, integrating both multi-sensor data acquisition and lightweight machine learning algorithm into a computation-limited wearable device. A one-dimensional CNN model, with a motion status recognition accuracy of 99%, is constructed and optimized for lightweight deployment on the wearable device, used for real-time data processing of multi-channel foot pressure and 3-axis acceleration signals. To further reduce the system power consumption, an event-triggered mechanism is developed based on human motion characteristics. The system ultimately realizes the low-power intelligent perception of five motion states, achieving an optimized power consumption of 6.3 mA in the active mode and 652 µA in the low-power mode. This study provides a potential solution for real-time intelligent monitoring of remote personalized healthcare.