A Non-Contact Sleep Posture Sensing Strategy Considering Three Dimensional Human Body Models

Sleep posture detection is very important in assisted living, healthcare, etc. With the rapid development of photoelectric technology, the cost of depth sensing equipment is decreasing; therefore, the depth images can be obtained in more ways. Then, SIFT features which represent the visual features of shapes are acquired from the images, feature will be mapped to some words and the bag-of-words paradigm is generated. In this work, a novel framework is proposed to detect various sleep postures. 3D scanned human body models are used as the input of the algorithm. Three subjects are considered in the experiment. Finally, PBHIKSVM is applied to detect sleep postures. The accuracy for classification is 92.5%, which is high enough for sleep state monitoring.