Improving Human Activity Recognition by Smart Windowing and Spatio-Temporal Feature Analysis

Abstract

This paper presents a promising approach to enhance multi-sensor based activity recognition in smart homes. The research is originated in the domain of Active and Assisted Living which mainly is about supporting older people to master their daily life activities. The paper proposes (a) a windowing technique which can be used for online sensor streaming and (b) a set of different statistical spatio-temporal features to recognize activities in real time. In order to check the overall performance, this approach was tested using the CASAS dataset. The results proved a high performance based on different evaluation metrics despite a large number of classes.