In current care facilities, bed-leaving sensors are used to prevent patients from falling at night. However, since these sensors normally detect the movement of a patient who has left the bed, when the sensor responds to the movement, the patient has already moved, and sometimes this means that he/she has already fallen from the bed. Since it is theoretically impossible to use the existing sensors for the purpose of preventing fall accidents, we focused on the relationship between sleeping positions and awakening timings. It is well known that people change posture frequently while sleeping. In this study, we focused on the frequency of change in sleeping posture, in order to verify which postures closely related to awakening timings. In consideration of the privacy of the care recipient, we have studied and developed a method to detect changes in posture while sleeping by deep learning technology using data obtained from a sheet-type pressuresensitive sensor.