Study on Human Behavior Classification by Using High-Performance Shoes Equipped with Pneumatic Actuators

Abstract

In Japan, accidents involving the falling of elderly people are increasingly becoming a problem. To solve this problem, walking training is effective for preventing falls of elderly people. In this study, a walking training system was developed in which high-performance shoes are used to improve the efficiency of walking training. The high-performance shoes have three functions: 1) measurement of plantar pressure using changes in the inner pressure of the insole, 2) leg movement measurement using a six-axis motion sensor, and 3) applying stimulus to the sole of the foot by changing the shape of the insole. A unique rubber element was developed for these functions. Furthermore, a system to predict the behavior of patients during walking training was developed. Based on experimental results, four types of behavior of patients during walking training were predicted. Moreover, leave-one-person-out cross validation was performed by the random forest (RF) machine-learning algorithm, and the F-measure was calculated. As a result, the four types of behavior were classified with an F-measure of 78.6%.