Ada : context-sensitive context-sensing on mobile devices

Abstract

This thesis describes the design, implementation, and evaluation of Ada, a context-sensing service for mobile devices. Ada explores new points in the accuracy-energy-responsiveness design space for mobile context sensing. The service exports an API that allows a client to express interest in one or more context types (mode-of-movement, indoor/outdoor, and entry/exit to/from named regions), and subscribe to specific modes within each context (e.g., "walking" or "running", but not any other movement mode). Each context type in Ada can be in one of a set of mutually exclusive states. Each context has a detector that returns its estimate of the mode. To achieve high accuracy and low energy consumption, the detectors take both the existing context and the desired subscriptions into account, adjusting both the types of sensors and the sampling rates. To accurately determine the movement mode, Ada uses a new peak frequency feature from acceleration magnitudes, combining it with two other features. We present results from trace-driven experiments over carefully labeled data from real users, finding that our mode-of-movement detector achieves an accuracy of 93%, out-performing previous proposals like UCLA (55%), EEMSS (83%) and SociableSense (72%), while consuming between 2 and 3x less energy.