Real-time Sleeping Posture Recognition For Smart Hospital Beds

Abstract

Unsuitable sleeping positions are the important contributors that result in bad sleep quality and even serious long-term consequences. Many studies emphasize that pressure sensor-based solutions are effective on the in-bed postures assessment in both home and hospital environments. Surprisingly, none of the studies considers Edge computing-based solution for body pose recognition on smart hospital beds. In this paper, we propose the development of a real-time sleeping posture recognition algorithm which is a combination of a preprocessing technique and an EfficientNet B0 based classifier with an AM-Softmax loss function. Experimental results confirm that our proposed method can gain the accuracy of over 99 % in 5-fold as well as 10-fold cross-validation and 95.32% in the Leave-One-Subject-Out (LOSO) validation for 17 sleeping postures, which greatly surpasses the previous method in the same task. Furthermore, our solution can satisfy the real-time requirement for various data sampling rates when deploying on the Edge computing-based smart hospital bed.