Health status along with assistive support requirements can be assessed by measures of activities of daily living. Advances in pervasive sensing and intelligent reasoning pave a way to monitor, i.e. detect and recognise, activities automatically and unobtrusively. The first task in monitoring activities is to detect when an activity has taken place based on a time series of sensor activation events. Inspired by the concepts of dynamic time warping and neighborhood counting matrix in similarity measures, this paper proposes a novel method to segment streams of sensor events for activity detection. Sensor segments may then be used as inputs to evidential ontology networks of activities for activity recognition.