Towards accelerometry based static posture identification

Human activity classification has wide-spread applications ranging from human computer interaction to disease progression studies. In this paper we propose a body posture model based on the Euler angles of the torso, arms and legs. The Euler angles are computed based on data streams originating from a wireless Body Sensor Network (BSN) comprising of nine accelerometers. Thereafter they are used to reconstruct different body postures based on an unsupervised learning and clustering algorithm. We validate our algorithm by implementing a classification engine in Matlab, capable of classifying subtle changes in posture with 97% accuracy.