Features Visualization of Young and elderly Gait Using Generative Adversarial Networks

Abstract

For gait guidance to elderly, previous studies have compared the gait of young adults and elderly, but only discrete assessments such as averages, and representative values have been carried out. To address this issue, evaluations have been carried out using principal component analysis on time series data from a certain gait cycle, but it is difficult to specify specific differences in joint angles. Furthermore, most of these studies assume that the walking ability of elderly is lower than that of young adults, and few studies have focused on the gait of active seniors, who can still walk briskly. Therefore, in this study, we focused on a technique called CycleGAN, a type of adversarial generative network, to visualize the time-series differences between the gait features of young adults and active seniors to support elderly in walking. As a result, the proposed method successfully generated and visualized reasonable gait feature generation for both gait types.