Transition Detection and Activity Classification from Wearable Sensors Using Singular Spectrum Analysis

Abstract

This paper proposes the use of singular spectrum analysis (SSA) to segment and classify human activities in real time by using an ear-worn Activity Recognition (e-AR) sensor. A similarity measure is calculated using SSA to construct a 3D feature vector from the 3 axes of e-AR signal. An algorithm based on the concept of clustering and buffering is then implemented in order to detect activity transition in real time as subjects perform their daily activities. An incremental subspace learning algorithm based on SSA is also proposed for activity classification. The proposed algorithm is applied to a group of five subjects performing daily activities and the results have shown the effectiveness of the method for transition detection and activity classification.