Automatic sleep database generation using frame subtraction and machine learning

Polysomnography is now a gold standard of sleep studies. However, the placement of sensors can be burdensome for both patients and hospitals. To solve this problem, some researchers have attempted to identify sleep-related disorders from video data. In these studies, videos were checked visually, which is time-consuming. Therefore, this study proposes a database generation method for sleep research using the frame subtraction method and image classification by machine learning. In this paper, we experimented with creating a body movement database using the proposed method. Accuracy validation results showed that body movement was classified with an accuracy of 0.98, precision of 0.80, recall of 0.67, and F-score of 0.73.