Sleep, which accounts for 1/3 of life, is a measure of health. Sleep disorders can cause a strain on the body and lead to drowsy driving accidents or industrial accidents. Both sleep time and quality of sleep are important, and everyone needs to sleep enough time deeply. Sleep patterns depend on genetic factors and vary from person to person. Therefore, it is necessary to analyze whether individual sleep patterns ensure a good night's sleep. In this paper, we have proposed a system of analyzing sleep patterns based on multi-sensor fusion including data-level fusion, feature-level fusion, and decision-level fusion. The proposed system determines whether to sleep well through the analysis of individual's sleep patterns. Based on this, we try to propose a customized environment configuration for a good night's sleep. Information about sleep status, such as the degree of tossing, snoring, and the one's body temperature, is directly related to the user. On the other hand, indoor temperature, humidity, illuminance, carbon dioxide (CO2) concentration, and ambient noise affecting deep sleep are information about the surrounding environment. The method of acquiring sleep status information is intended to be implemented in an unrestrained and unconscious state considering that the user is in a sleep state. That is, a smart pillow with built-in pressure sensors determines the torsion degree, and a non-contact temperature sensor attached to ceiling measures the body temperature. In order to increase the accuracy of sleep pattern analysis, we propose a bottom-up fusion process in which the analysis result of the decision-level fusion controls the data-level fusion and feature-level fusion. As the results of applying the integrated fusion process for actual subjects, three sleep patterns were derived: deep sleep, intermittent sleep, and sleep disorder. In addition, information for controlling sleep environments according to sleep patterns was generated. The proposed approach will generate more detailed sleep conditions and information on controlling sleep conditions for each individual by expanding the number of subjects.
