Abstract:
Smart home is a promising solution for the aging population who requires assistance but prefers living independently at home. Smart home is a manifestation of pervasive computing which incorporates multimodal sensors, actuators, devices, information, and communication technologies to gather different information about the environment and its users. One of the main characteristics of pervasive computing is context awareness. In this thesis, we present the development of a context-aware activity recognition system using wearable and sensors embedded in environment to continuously monitoring activities of daily living of elderly people. The proposed system fuses contextual information about user’s physical activity, location and interactions with objects in the environment using ontology, and accurately recognize the activities. To develop a robust physical activity recognition, we propose a novel adaptive sliding window segmentation to select a more effective window segmentation of acceleration signals. In addition, we propose an activity transition diagram to be integrated into the activity classification algorithm to validate the activity transition after window classification. To overcome the limitation of ontology in dealing with uncertainty due to missing sensor data, we propose a novel reasoning algorithm that integrates ontological reasoning mechanism with Dempster-Shafer theory of evidence. The algorithm provides support for handling uncertainty by quantifying uncertainty while aggregating contextual information and produce a degree of belief to facilitate a more robust decision making in activity recognition. To further enhance the recognition accuracy of the system, we present the integration of user context into the activity ontology to handle uncertainty due to missing sensor data. In addition, the approach allows additional and more precise inference of activities and recognizes activities that do not involve interaction with objects in the environment.