Study on gait discrimination method by deep learning for biofeedback training optimized for individuals

Abstract

In this research, to develop a biofeedback training system where trainees can efficiently train inadequacies that do not satisfy ideal walking using a deep learning, we examine a method that discriminates between ideal walking and nonideal walking. In the experiment, to examine the walking components used for the input data, the ground reaction force and joint angle were measured when young people walked normally and when they walked with a brace, to simulate elderly motions. Further, these data were discriminated between conditions as input data using a Convolution Neural Network (CNN). The average accuracy was 79.5% when all walking components were used as input data. In addition, it is thought that it is most suitable to discriminate walking by using all walking components, in consideration of implementation in the system.