Classification of walking patterns on inclined surfaces from accelerometry data

Abstract

This paper describes the classification of walking patterns on ascending and descending slopes based on features extracted from data recorded using a single waist-mounted tri-axial accelerometer. A 19-dimensional set of salient features representing the hill walking patterns were obtained based on gait cycle analysis related to the acceleration data in the anterior-posterior (AP), medio-lateral (ML), and vertical (V) directions. A Gaussian mixture model (GMM) classifier was used to perform a four way classification task, discriminating between two inclines and two declines. An overall classification accuracy of 90.9% was achieved for the four different human gait patterns referring to four different paved gradients (up or down 4.8% and 17.3% gradients).