A windowing approach for activity recognition in sensor data streams
Fadi Al Machot, Heinrich C. Mayr, Suneth Ranasinghe · 2016
Determining the appropriate data window size for online sensor data streams to recognize a specific activity is still a challenging task. In particular, when new sensor events are recorded. This paper proposes a windowing algorithm which presents promising results to recognize complex activities, e.g., in a smart home environment. The underlying basic idea is to analyze the sensor data in order to identify the set of “best fitting sensors”: it contains those sensors that most contribute to the recognition task, and therefore should be considered in a window. To validate our approach, we applied it on the CASAS data set which is an international data set for activity recognition. Based on the promising results, we believe that this algorithm can assist to detect human activities. Thus, our approach might be used in Active and Assisted Living Environments (AAL), where activity recognition is required to distinguish the types of help, a person needs to master his/her daily life activities.