Application of Machine Learning Technology to Gait Analysis and Training

Abstract

In super-aging Japan, as a measure against increasing the number of people requiring long-term care, effective gait training is more necessary than guidance by physical therapists. As a solution, there are several training methods that measure information on gait and feed the information with the target value back to the trainee in real time. Gait training for stroke patients has been proven to effectively improve the lateral symmetry. However, when challenges to the gait of an individual are unclear, such as with active seniors, a setting of the target values becomes an issue. As such, we used a machine learning technique, i.e., a multi-channel deep convolutional neural network, and developed a method to generate the target values that satisfy the ideal gait characteristics, and a gait feedback training system that considers the individual physical differences in the trainees by using this method. In this chapter, we discuss previous studies on the gait information feedback training system. We describe basic knowledge and the development of technology regarding a neural network, which is a representative machine learning method, and discuss previous studies on machine learning for a gait analysis. In addition, in this chapter, we describe the proposed gait feedback training system and a method for setting the training target, along with a verification of its effectiveness when the training target is a propensity to stumble.