Human Sleeping Posture Recognition Based on Graph Neural Network Using Sensors Combined With Airbag Bed Detection

Approximately one-third of a human’s lifetime is spent sleeping, and sleeping posture plays a pivotal role in both sleep quality and overall health. Therefore, there is a pressing need for an affordable, yet high-performance sleeping posture recognition system with adaptive adjustment capabilities. In this study, we present a sleeping posture recognition system based on a graph neural network (GNN), utilizing an airbag mattress integrated with sensors. The system positions airbags at five key body regions to form a mattress and transmits the real-time data to a PC via an air pressure sensors module. The data is processed using three different graph topologies and fed into a GNN-based model, which incorporates two task types (node-level or graph-level) for classifying six health-related postures, achieving a recognition accuracy of up to 98%. Furthermore, the system employs a closed-loop control mechanism to optimize support distribution via real-time pressure monitoring and local air pressure adaptation, thereby enhancing sleep quality.