Normal sleep is characterized by the presence of several major posture shifts as well as some smaller body movements that do not necessarily have an effect on sleeping posture. The main ergonomic function of sleeping systems, to provide proper body support, requires the knowledge of the adopted postures throughout the night. This can be achieved by accommodating bedding systems with sensors that measure the deformation of the system due to the human body on top of it. An automatic posture recognition algorithm based on such bed measurements is developed and validated with independent video analysis. The algorithm uses decision rules that are constructed by support vector machines, a kind of learning machine that calculates the optimal separating hyperplane between two classes of a training set of feature vectors. Results are promising with a mean correspondence of 0.92 between automatic and validation posture scorings. Apart from its ergonomic relevance, the developed posture recognition can also serve as a base to compute other motion related parameters that might be relevant to relate to sleep quality.
Vincent Verhaert Bart Haex Tom De Wilde Raymond Cluydts Daniel Berckmans Johan Verbraecken Jos Vander Sloten