Semi-supervised segmentation for activity recognition with Multiple Eigenspaces

Body Sensor Networks (BSNs) are increasingly being used in pervasive sensing environments including healthcare, sports, wellbeing, and gaming. Activity segmentation using BSN is challenging and the use of manual annotation is subjective and error prone. In this paper, we investigate a semi-supervised activity segmentation method using a Multiple Eigenspace (MES) technique based on Principal Components Analysis (PCA). Results show that the method can reliably perform activity segmentation and the classification results based on HMMs demonstrate the practical value of the proposed technique.