Sleeping Posture Classification Through a Multi-Sensing Smart Mattress Based on Fiber Bragg Grating Sensors: A Feasibility Study

Monitoring sleeping posture plays a pivotal role in the diagnosis and prevention of sleep-related disorders. Artificial intelligence and unobtrusive devices are commonly used to track sleep biomarkers such as posture. Among several devices, smart mattresses are gaining momentum in this field. Fiber Bragg grating sensors (FBGs) are gaining broad acceptance for mattress instrumentation. In this study, we propose the implementation of machine learning and deep learning classification models (Support Vector Machine -SVM-, Decision Tree -DT-, Long Short-Term Memory -LSTM- and 2D Convolutional Neural Network -2D CNN-) for automatic posture recognition through a multi-sensing FBG-based mattress. Experimental testing on five healthy volunteers (both male and female) taking different sleeping postures (i.e., supine, right lateral, left lateral, and prone) allowed for the assessment of the proposed devices' performance under different sleep-related conditions. In particular, we carried out a binary classification (axial vs lateral postures and left vs right lateral postures).The results demonstrated higher accuracy in the case of axial vs lateral task (up to 89%) than right lateral vs left lateral task (i.e., 80%), employing the use of 2D CNN.