Pocket, Bag, Hand, etc. - Automatically Detecting Phone Context through Discovery

Most top end smart phones come with a handful of sensors today. We see this growth continuing over the next decade with an explosion of new distributed sensor applications supporting both personal sensing with local use (e.g., healthcare) to distributed sensing with large scale community (e.g., air quality, stress levels and well being), population and global use. One fundamental building block for distributed sensing systems on mobile phones is the automatic detection of accurate, robust and low-cost phone sensing context ; that is, the position of the phone carried by a person (e.g., in the pocket, in the hand, inside a backpack, on the hip, arm mounted, etc.) in relation to the event being sensed. Mobile phones carried by people may have many different sensing contexts that limit the use of a sensor, for example: an air-quality sensor offers poor sensing quality buried in a person’s backpack. We present the preliminary design, implementation, and evaluation of Discovery, a framework to automatically detect the phone sensing context in a robust, accurate and low-cost manner, as people move about in their everyday lives. The initial system implements a set of sophisticated inference models that include Gaussian Mixture Model and Support Vector Machine on the Nokia N95 and Apple iPhone with focus on a limited set of sensors and contexts. Initial results indicate this is a promising approach to provide phone sensing context on mobile phones.