Stream data analysis of body sensors for sleep posture monitoring: An automatic labelling approach

Abstract

Sleeping is one of the most important activities in our daily lives. However, very few people really understand their sleeping habits, which affect sleep-related diseases such as sleep apnea, back problems or even snoring. Most current techniques that monitor, predict and quantify sleep postures are limited to use in hospitals and/or need the intervention of caregivers. In this paper, we describe a system to automatically monitor, predict and quantify sleep postures that may be self-applied by the general public even in a non-hospital environment such as at a persons home. A Random Forest approach is adopted during training to predict and quantify sleep postures. After going through training procedures, a person needs only one sensor placed on the wrist to recognize the persons sleep postures. Our preliminary experiments using a set of testing data show about 90 percent accuracy, indicating that this design has a promising future to accurately analyze, predict and quantify human sleep postures.