Recognizing User's Context from Wearable Sensors: Baseline System
Brian Clarkson, Kenji Mase, Alex Pentland · 2000
INTRODUCTION We describe a baseline system for training and classifying natural situations. It is a baseline system because it will provide the reference implementation of the context classifier against which we can compare more sophisticated machine learning techniques. It should be understood that this system is a precursor to a system for understanding all types of observable context not just location. We are less interested in obtaining high precision and recall rates than we are in obtaining appropriate model structures for doing higher order tasks like clustering and prediction on a user's life activities. II. BACKGROUND There has been some excellent work on recognizing various kinds of user situations via wearable sensors. Starner [6] uses HMMs and omnidirectional and directional cameras to determine the user's location in a building and current action during a physical game. Aoki also uses a head mounted directional camera to d