Feature Selection and Activity Recognition from Wearable Sensors
We describe our data collection and results on activity recognition with wearable, coin-sized sensor devices. The devices were attached to four different parts of the body: right thigh and wrist, left wrist and to a necklace on 13 different testees. In this experiment, data was from 17 daily life examples from male and female subjects. Features were calculated from triaxial accelerometer and heart rate data within different sized time windows. The best features were selected with forward-backward sequential search algorithm. Interestingly, acceleration mean values from the necklace were selected as important features. Two classifiers (multilayer perceptrons and kNN classifiers) were tested for activity recognition, and the best result (90.61 % aggregate recognition rate for 4-fold cross validation) was achieved with a kNN classifier.
Unable to display preview. Download preview PDF.
- 3.Fujinami, K., Pirttikangas, S., Nakajima, T.: Who opened the door?: Towards the implicit user identification for sentient artefacts. In: Fishkin, K.P., Schiele, B., Nixon, P., Quigley, A. (eds.) PERVASIVE 2006. LNCS, vol. 3968, pp. 107–111. Springer, Heidelberg (2006)Google Scholar
- 5.Huynh, T., Schiele, B.: Analyzing features for activity recognition. In: Proc. Joint Conf. Smart Objects and Ambient Intelligence (sOc-EUSAI 2005), pp. 159–163 (2005)Google Scholar
- 6.Kukolich, L., Lippmann, R.: Lnknet user’s guide (August 1999) (Available, 18.4.2006), http://www.ll.mit.edu/IST/lnknet/
- 8.Yamabe, T., Takagi, A., Nakajima, T.: Citron: A context information acquisition framework for personal devices. In: Proc. the 11th IEEE Int. Conf. Embedded and Real-Time Computing Systems and Applications (RTCSA 2005), pp. 489–495 (2005)Google Scholar