Improving human activity recognition using subspace clustering

Abstract

Activity recognition attracted much interest in pervasive sensing due to extensive application in human daily life from health monitoring to security monitoring. It utilizes collection of data from low level sensor to learn about human behaviors and activities, so that services can be provided by function of detecting anomalies, remote interventions or prompts. The approach of human activity modeling and recognition still confronted with a challenge on issues of modeling human activity in human perspective. However, the traditional learning-based approaches are not sufficient to capture the characteristics of human activity because they still use traditional clustering method to process sensor data which consists of multidimensional information. This paper describes a subspace clustering-based approach to recognize human activity and detect exceptional activities. Different from many approaches, the proposed approach use subspace clustering based approach to model of human activity in order to improve accuracy of activity recognition. Finally, the proposed approach has been validated on data collected from RFID-based systems, which results demonstrate the effectiveness of the proposed improvents.