| The future of human computer interaction systems lies in how intelligently these systems can take into account the user's context, that is, how well the data that they produce characterizes the user's current situation. Context awareness is essential for ubiquitous and wearable computing. Research on recognizing the daily activities of people has progressed steadily, but little focus has been devoted to recognizing activities along with the movements involved in it. For many applications such as rehabilitation, sports medicine, geriatric care, and health/fitness monitoring, the importance of combined recognition of activity and movements within an activity can drive health care outcomes. Motion recognition aims at recognizing the actions of one or more users from a series of observations on the users' actions and environmental conditions.;Sensor-based motion recognition integrates the emerging area of wireless sensor networks with novel machine learning techniques to model a wide range of human motions. A novel algorithm is proposed that can be tuned to recognize on-the-fly either range of activities or fine motor movements within a specific activity using wirelessly connected sensor motes (equipped with accelerometers and gyroscopes) attached to different body sites. This thesis describes a novel algorithm for both situations and also presents a case study on optimal feature set from sensor values and various parameter values for the algorithm to detect the fine motor movements within an activity. |
