ROC Analysis of Partitioning Method for Activity Recognition Using Two Microphones

We present an analysis, using ROC curves, of a method for partitioning continuous activity data using two microphones. The algorithm is based on utilising the difference in sound intensity recorded by microphones placed on the upper arm and on the wrist. We show that the method is feasible for detecting activities involving the interaction of the hand with tools or machinery where noise is produced close to the hand. We also show that the method is relatively robust across multiple subjects.