S1601-2-1 腕の動き計測を用いた行動判別手法の高精度化(生物医学工学における計測と制御(2))

Abstract

Recently developed classification schemes of human activities deal with dynamic activities and therefore are not adequate to classify static activities such as eating and sitting, which are necessary to monitor an entire daily life pattern. In this paper, we propose a classification method of human activities that can be applied to both static and dynamic activities. Usually tri-axial accelerometer data is used to classify human activities, but sitting position and standing position could not be classified clearly by only accelerometer. We use tri-axial angular velocity data in addition to accelerometer data, for the classification of static activities. We found that the choice of the part of human body onto which the motion sensors were placed has an effect on the classification performance, the wrist being the best position to improve classification accuracy of static activities. Our classification scheme uses the k-nearest neighbors algorithm on selected features extracted from tri-axial accelerometer data and tri-axial angular velocity data.