Unobtrusive Sleep Posture Detection Using a Smart Bed Mattress with Optimally Distributed Triaxial Accelerometer...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Science

Background:

• Accurate sleep posture detection is crucial for assessing sleep quality and monitoring health.
• Improper sleep postures can lead to musculoskeletal issues, respiratory problems, and exacerbate conditions like sleep apnea.
• Existing methods like wearable sensors, cameras, and pressure mats have limitations including discomfort, privacy concerns, and high costs.

Purpose of the Study:

• To develop a low-cost, non-contact system for effective sleep posture detection.
• To improve upon traditional methods by offering a comfortable and privacy-preserving solution.
• To validate the system's accuracy in identifying various sleep postures.

Main Methods:

• Utilized eight triaxial accelerometers for non-contact data acquisition.
• Employed an improved density peak clustering algorithm with K-nearest neighbor for classification.
• Developed a Parallel Convolutional Spatiotemporal Network (PCSN) integrating CNN, LSTM, and Bi-LSTM modules.

Main Results:

• The PCSN accurately distinguished six sleep postures (prone, supine, left log, left fetus, right log, right fetus) with an average accuracy of 98.42%.
• The system outperformed state-of-the-art deep learning models across all metrics, achieving 98.64% precision, 98.18% recall, and 98.10% F1 score.
• Demonstrated the system's effectiveness in real-world sleep posture identification.

Conclusions:

• The developed low-cost, non-contact system offers a promising solution for sleep posture detection.
• The PCSN deep learning model shows high accuracy and robustness in classifying sleep postures.
• The system has significant potential for applications in sleep studies, health monitoring, and the prevention of conditions like pressure ulcers and sleep apnea.