FMCW Radar-based Sleep Posture Monitoring Through Logic and Deep Learning Methods.

In this study, we introduce a non-invasive approach for sleep posture monitoring based on Frequency-Modulated Continuous Wave (FMCW) radar. In contrast to prior research, our method employs a novel logic-based approach to analyze sleep postures during the entire sleep process using only one radar device. The inspiration behind our proposal is the observations made regarding the downsampling calculation, which allows the Range, Doppler, Azimuth, and Elevation-Time (RT, DT, AT, and ET) map to focus more on breathing movement. Subsequently, a logical approach is proposed based on these data to recognize supine, prone, and side postures, and then further separate side postures into left and right postures. In addition, a custom Convolutional Neural Network (CNN) model is designed to compare the results with the performance of the logic-based method. Our evaluation was conducted using data collected from 10 subjects, which showed promising results. The logic-based method demonstrates high accuracy, achieving an overall accuracy of 98.11% for 3category classification and 94.79% for 4-category classification. For the CNN-based approach, our model attains an accuracy of 93.25% in 4-category classification under subject-independent evaluation. Compared to the CNN-based method, the logic-based method provides better accuracy, but the CNN-based approach has the advantage of real-time applicability. These results underscore the effectiveness of our non-invasive FMCW radar-based sleep posture monitoring system.