Sleep quality and quantity is an important factor in the prevention of fraitly and serious deseases such as depression and diabetes. Therefore the devices that allow sleep monitoring are studied for well being. Body movement during sleep can be related to particular desesases such a sleep apnea and restless legs syndrome. In this paper we present a study on movements detection during sleep and the estimation of inactivity duration. Contactless thermal sensor (thermopile) is use to detect thermal variation caused by body movement in bed. A new adaptive segmentation method including symbolic representation (SAX) is introduced and implemented to compare identified sleep phases by our system to a reference hipnogram. The proposed sensor and his processing algorithm has been evaluated on 11 subjects. This evaluation shows good correlation between automatic segmentation and the hypnogram reference labels.