SPAR: Service-based Personal Activity Recognition for Mobile Phones

Smart phones have become powerful platforms for mobile communication and applications. This paper presents basic technology that will enable the phone to extend such applications with context awareness under realistic conditions. Recognition is carried out by a service-based context recognition architecture which creates an evolving classification system based on feedback from the user community. The approach uses classifiers based on fuzzy inference systems which use live annotation to personalize the classifier instance on the device to the its user. Our recognition system is designed for everyday use: it is independent of placement (no assumed or fixed position), requires only very little (1-3 minutes per activity) personalization effort from the user and can detect a high number of activities. The results demonstrate the ability of the system to use personalization and the user community as forces for optimization, achieving classification rates upwards of 97% for 10 classes in an evaluation with 20 users and over 500 minutes of data.