Real-Time Sensor Observation Segmentation For Complex Activity Recognition Within Smart Environments

Activity Recognition (AR) is at the heart of any types of assistive living systems. One of the key challenges faced in AR is segmentation of the sensor events when inhabitant performs simple or composite activities of daily living (ADLs). In addition, each inhabitant may follow a particular ritual or a tradition in performing different ADLs and their patterns may change overtime. Many recent studies apply methods to segment and recognise generic ADLs performed in a composite manner. However, little has been explored in semantically distinguishing individual sensor events and directly passing it to the relevant ongoing/new atomic activities. This paper proposes to use the ontological model to capture generic knowledge of ADLs and methods which also takes inhabitant-specific preferences into considerations when segmenting sensor events. The system implementation was developed, deployed and evaluated against 84 use case scenarios. The result suggests that all sensor events were adequately segmented with 98% accuracy and the average classification time of 3971ms and 62183ms for single and composite ADL scenarios were recorded, respectively.