Motion analysis by wireless sensor shoes using machine learning

Abstract

In recent years, researches have been conducted to recognize human behavior using wearable sensors for the purpose of health management, work analysis, safety and security, etc. The behavior is so complicated that it has been studied from various aspects such as engineering, medical science or biology. In this study, we constructed a system that automatically recognizes motion by using sensor shoes with multiple 3-axis tactile sensors arranged on the insole in combination with supervised machine learning. We measured the sole forces during the five motion of walking, climbing and descending stairs, going up and down a slope, and built a supervised machine learning model based on those data, and then input test data with unknown behavior to this model. Motion identification and accuracy evaluation were carried out. As a result of the experiment, the identification accuracy of the two types of motions was from 63% to 93%.