Activity recognition plays an important role in smart homes for aged-care. In this paper, we formulate the problem of activity recognition and propose a new method based on spatial-temporal constraints to carry out activity recognition, which consists of five phases: initialisation, segmentation, sensor data representation, activity exploration as well as activity identification. Besides, we analyse the time complexity and space complexity of our approach in theory. To evaluate our approach, we carried out experiments on real dataset from Wireless and Mobile Network Laboratory, Tamkang University. The experimental results demonstrate an improvement of 5.6% in the accuracy on average of recognised activities in comparison to the method of support vector machine (SVM).
International Journal of Ad Hoc and Ubiquitous Computing 33(3), p.168-183