Abstract

This paper presents details of a convenient andunobtrusive system for monitoring daily activities. A smartphone equipped with an embedded 3D-accelerometer was wornon the belt for the purposes of data recording. Once collectedthe data was processed to identify 6 activities offline (walking,posture transition, gentle motion, standing, sitting and lying).The processing technique adopted a novel hierarchicalclassification. In the first instance, rule-based reasoning is usedto discriminate between motion and motionless activities.Following this the classification process utilizes two multiclassSVM (support vector machines) classifiers to classify themotion and motionless activities, respectively. The classifierswere trained on data from one subject and tested on 10subjects. The experiments demonstrate that the hierarchicalmethod can reduce misclassification between motion andmotionless activities. The average accuracy was improvedcompared with using a single classifier by using thisclassification method (82.8% vs. 63.8%), and is important forproviding appropriate feedback in free living applications.