Recognition of a variety of activities considering smartphone positions

Abstract

We present a high-accuracy recognition method for various activities using smartphone sensors based on device positions. Many researchers have attempted to estimate various activities, particularly using sensors such as the built-in accelerometer of a smartphone. Considerable research has been conducted under conditions such as placing a smartphone in a trouser pocket; however, few have focused on the changing context and influence of the smartphone position. Herein, we present a method for recognising seven types of activities considering three smartphone positions, and conducted two experiments to estimate each activity and identify the actual state under continuous movement at a university campus. The results indicate that the seven states can be classified with an average accuracy of 98.53% for three different smartphone positions. We also correctly identified these activities with 91.66% accuracy. Using our method, we can create practical services such as healthcare applications with a high degree of accuracy.