A Sleep Posture Detection Method for Pregnant Women Based on AI Algorithms and a Flexible Sleep Monitoring Belt with a MEMS IMU

This paper proposes a new method for real-time detection of pregnant women's sleep postures, namely a sleep posture detection system using a flexible sleep monitoring belt equipped with a MEMS IMU (inertial measurement unit) and a machine learning algorithm. The system provides non-invasive, continuous, and high-precision monitoring to address key challenges in maternal health. By analyzing the changes in acceleration and angular velocity data captured by the IMU, we propose a multi-stage processing flow, including wavelet packet transform for denoising, feature extraction using short-time energy analysis, and classification through an integrated machine learning model. Experimental results show that the KDTREE model achieves a 100% accuracy, outperforming other models such as CNN and BP networks. This research helps improve sleep quality monitoring during pregnancy through low-cost, non-wearable solutions.