Our prototype uses only two wireless motion sensor nodes placed on the waist and on the thigh of wearer, and a personal smartphone running the end-user application which is able to detect four basic activities (lying down, sitting, standing, and walking). This is achieved with or without an individual training phase, and with an overall average accuracy of about 98%.
The application also provides an incremental learning system that users may use to increase the set of activities that can be recognized (e.g., to add a “kicking” or “jumping” activity).
R. Giannantonio, R. Gravina, P. Kuryloski, V.-P. Sepp¨a, F. Bellifemine, J. Hyttinen, and M. Sgroi, Performance analysis of an activity monitoring system using the SPINE framework, 3rd Int. Conf. Pervasive Comput. Technol. Healthc., Apr. 2009, pp. 1–8.