Real-time Context Recognition

The aim of wearable computing is to build intelligent machines that provide automatic and autonomous support in people's everyday lives. Accurate and comprehensive recognition of a user’s context is an important step towards that goal. In this thesis a light-weight PDA-based wearable system is presented that classifies in real-time location, speech, posture and a number of activities, based on audio and acceleration sensing. After a pre-classification step, which is focused on the four categories, a common sense model is used to improve overall recognition accuracy in a post-classification step. The context information won is used in a diary application to trigger picture taking and audio recording of interesting moments of the wearer’s life