Covered body analysis in application to patient monitoring

Abstract

Patient monitoring in medical applications such as diagnosis of sleep disorders commonly adopts invasive monitoring equipments such as pulse oximetry and polysomnogram (PSG), but their attachment to the patient's body disturb sleep and therefore compromise results. Furthermore, the invasive approaches often fail to monitor continuously because the devices can be pulled off by the subject during sleep unconsciously. This paper presents an automated noninvasive video monitoring approach to analyze (covered) human activity in conditions with persistent heavy occlusion. The proposed method is a model-based approach, employing both static shape features and dynamic motion features to suppress false positive detection, to identify human activity, and to self-improve the covered human pose estimation.