A motion sequence fusion technique based on PCA for activity analysis in body sensor networks

Human movement analysis by means of mobile sensory platforms is an ever-growing area with promise to revolutionize delivery of healthcare services. An effective data fusion technique is essential for understanding the inertial information obtained from distributed sensor nodes. In this paper, we develop a data fusion model based on the concept of principal component analysis. Unlike traditional fusion techniques which deal with statistical feature space, our model operates on motion transcripts, where each movement is represented as a sequence of basic building blocks called primitives. We describe how our model transforms transcripts of different nodes into a unified transcript by integrating the most relevant primitives of movements. Finally, we demonstrate the performance of our transcript fusion model for action recognition using real data collected from three subjects.