Sleep Monitoring Based On Contactless Perception

Sleep is a significant part of human life activities.The quality of sleep is closely related to people’s physical and mental health.With the increasing pressure of life and the increasing degree of social aging,more and more people have sleep disorders.Thus,sleep health monitoring becomes more and more important.However,sleep diagnosis in medical places requires professional equipment and personnel,which is expensive and uncomfortable.Therefore,it is not suitable for long-term monitoring and monitoring at home.With the rapid development of video image technology and deep learning technology,video-based remote monitoring method can effectively detect various vital signs.Compared with contact methods,non-contact sleep monitoring will be more comfortable.Since the changes of respiration and heart rate can reflect the sleep state and detect sleep disorders,this paper monitors the respiration,heart rate and sleep state of people in the dark environment at night based on the non-contact method of near-infrared video.The specific work is as follows:1)Aiming at the problems that the existing non-contact respiratory monitoring methods are time-consuming and dependent on ROI selection,a respiratory monitoring module based on super-pixel edge detection is designed.Firstly,only the local processing of the video frame instead of reconstructing the whole video can reduce the time complexity and meet the requirements of real-time.Secondly,super pixels are generated in regions of interest and the edge of super pixels is detected.Only the edge pixels rather than the pixels in the whole ROI are filtered and analyzed in time domain,which effectively avoids the influence of most flat areas in ROI on respiratory measurement.Finally,an adaptive dynamic adjustment mechanism is introduced to make the measurement results not affected by the natural shaking of the human body in the shooting process.Experiments show that the proposed breath detection method based on super-pixel edge detection can not only meet the requirements of real-time monitoring,but also effectively judge sleep-disordered breathing such as apnea.2)Aiming at the problems that the existing video heart rate detection methods are mostly based on RGB videos,which is easily affected by the environment,a nighttime heart rate detection module based on independent vector analysis is designed.Since the heart rate value in the sleep state will be at least 10 times / minute lower than that in the awake state,the sleep state can be judged by the change of heart rate value.Firstly,in order to eliminate the influence of lighting conditions,this paper measures heart rate based on near-infrared videos.Secondly,in order to make up for the information gap between near-infrared video and RGB video,multiple ROI are selected,and the method of independent vector analysis is adopted,and multiple candidate signals of heart rate are obtained through time-domain filtering and spectrum analysis.Finally,the dynamic programming algorithm is used to screen out the heart rate candidate signals that conform to the actual heart rate variation,and the abnormal heart rate values are corrected.Experiments show that the proposed heart rate detection method based on independent vector analysis not only has a good detection effect when the heart rate remains stable,but also successfully detects the change trend of heart rate when there is an obvious change trend.The sleep state can be judged according to the change trend of heart rate that decreases when falling asleep,remains stable during sleep and rises when waking up.3)Aiming at the situation that the respiration and heart rate detection methods based on face recognition fails when the facial image is incomplete,a sleep state detection module based on anomaly detection method is designed.During sleep,the human body has unconscious body movements,such as turning over,which will make the face blocked or even disappear in the video,resulting in the failure of breath and heart rate detection methods based on face detection.In this paper,the prediction-based anomaly detection network is used to judge the sleep state.Since people basically remain still when they are sleeping,the state of the limbs remaining still is regarded as falling asleep,otherwise it is regarded as not falling asleep.Experiments show that sleep state detection based on anomaly detection method can effectively distinguish different states.