link datasets https://physionet.org/content/pmd/1.0.0/
Sleep posture analysis is widely used for clinical patient monitoring and sleep studies. The correct and automatic detection of a patient and the way he sleeps can be very usefull to take care of him. In this study, the classification is done using data acquired by a commercial pressure map system from the PmatData dataset. This given data is preprocessed in order to reduce its dimension. Not only the classification accuracy of the model is taken into consideration but also the total amount of data that is needed to perform it. Two way of building a new more compact dataset are presented, the first one using just a Principal-Component-Analysis (PCA) over the data and the second one using also a K-means approach to simply stored some relevant points of the body shape of the subject. Moreover the classification is performed with a Convolutional-Neural-Network (CNN) in the first case and with a Fully-Connected-Neural- Network (FCNN) for the second one. Both models are capable of accurately detecting subjects and their sleeping postures. A tradeoff between the accuracy of the classification and the total amount of the data was searched. This can be usefull for some remote computation of the sleeping posture or for subject identification.