Rotations Recognition in Sleeping Humans by Means of Bioradiolocation and Machine Learning

Bioradiolocation is a contactless and unobtrusive method for respiration, heartbeat and movement parameters estimation. In this paper we present movement detection technique which allows to distinguish the whole body rotations from other movements in a sleeping human. We used radar recordings from 13 different subjects to train machine learning models and tested them on signals of other S volunteers. We also compared classification accuracy, sensitivity and specificity between one and two-radar experimental scheme. The best model proved to be Quadratic SVM with 91.67%, 95.24% and SS.10% for these measures of the performance respectively. This result may be useful in diagnosis of restless sleep disorder both in children and adults.