S. Omatu et al. (Eds.): IWANN 2009, Part II, LNCS 5518, pp. 796–799, 2009. © Springer-Verlag Berlin Heidelberg 2009 Activity Recognition from Accelerometer Data on a Mobile Phone Tomas Brezmes 1 , Juan-Luis Gorricho 2 , and Josep Cotrina 2 1 France Telecom R+D, Spain tomas.brezmes@orange-ftgroup.com 2 Departamento de Telemática, UPC, Spain {juanluis,jcotrina}@entel.upc.edu Abstract. Real-time monitoring of human movements can be easily envisaged as a useful tool for many purposes and future applications. This paper presents the implementation of a real-time classification system for some basic human movements using a conventional mobile phone equipped with an accelerometer. The aim of this study was to check the present capacity of conventional mobile phones to execute in real-time all the necessary pattern recognition algorithms to classify the corresponding human movements. No server processing data is involved in this approach, so the human monitoring is completely decentralized and only an additional software will be required to remotely report the human monitoring. The feasibility of this approach opens a new range of opportunities to develop new applications at a reasonable low-cost. Keywords: Pattern recognition, human movement’s detection, accelerometer. 1 Introduction An aging population is one of the main concerns of present administrations. Thinking about new health-care paradigms to diminish the expected ever increasing health-care budget is becoming a real necessity. Thankfully, recent progress in information com- munication technologies and sensor miniaturization have provided the foundation for the development of systems concerned with the remote supervision of home-based physiological monitoring. In particular, a real-time monitoring of human movements is expected to be a practical solution to monitor aged people or any human being who needs to be under medical control. At present there is an extended bibliography on the field of human movement’s de- tection with studies considering wearable sensor units [1]-[3] or employing multiple accelerometer units located on different body sites [4]-[7]. On the other hand, many studies are devoted to improve or compare the accuracy of pattern recognition to classify the human movements. The most commonly used techniques come from applying artificial intelligence principles: decision trees, k-nearest neighbors, neural networks, support vector machine, etc. [7]-[9]. This paper exposes the implementation of a real-time classification system for some representative human movements: walking, climbing-up stairs, climbing-down