Sleeping position detection based on multi-domain features of ballistocardiogram

Research shows that different sleeping positions can affect people's sleep quality. Poor sleep quality can easily lead to a variety of cardiovascular diseases. In order to better detect people's sleep conditions, this paper proposes a classification method based on ballistocardiogram (BCG). Non-contact optical fiber sensors were employed to capture BCG signals from the chest region of individuals in supine, prone, left lateral, and right lateral positions. Considering that the sleeping posture will cause the characteristic difference of the signal in time domain, frequency domain and phase, this study introduces a multi-domain feature extraction approach based on sinusoidal function fitting. This method enables the extraction of time-frequency phase features from the BCG waveform for the classification of the four sleeping postures. Experimental results demonstrate a fitting accuracy of 97% for the four-term sine function and an 83.54% accuracy for sleeping posture classification using multi-domain features.