Home // International Journal On Advances in Life Sciences, volume 12, numbers 3 and 4, 2020 // View article

Recognition Method for a Temporary Change in Walking based-on Anomaly Detection and Classification

Authors:
Shin Morishima
Akira Urashima
Tomoji Toriyama

Keywords: Walking recognition; Classification; Anomaly detection; Human activity recognition

Abstract:
In recent years, the global population of the elderly has increased, and one of the challenges faced by this population is the increased vulnerability to falls. Two approaches to alleviate this problem are determining actions that cause these falls and preventing falls using the detection results. A temporary change in walking (i.e., stumbling and staggering) is a typical cause of a fall. However, existing studies do not focus on a temporary change in walking; they only distinguish between walking or other activities and recognize walking speed. In this paper, we propose a method to detect the change in walking using change point detection (i.e., anomaly detection for time series data) and a classification method for the multiple types of change. Moreover, we assume four cases classified using available data, and we propose the parameter setting of the proposed method for each case to be applied in diverse scenarios. During the evaluation, four anomalous walking videos (where, anomaly represents a temporary change) are used. Thus, the accuracy of the anomaly detection of the proposed method is up to 93.5%, and the four types of detected anomalous walking are classified into three clusters in 89.1% of cases based on each characteristic.

Pages: 92 to 101

Copyright: Copyright (c) to authors, 2020. Used with permission.

Publication date: December 30, 2020

Published in: journal

ISSN: 1942-2660