Evaluating an In-Home Exercise Program Using Vision-Based Edge AI for Elderly Healthcare

In Japan, many in-home exercise programs have appeared to improve the motor function of the elderly through self-care. However, a challenging point always exists in evaluating the process of exercise programs quantitively and non-invasively for further reflecting the health status. As a preliminary progress, we present and discuss a vision-based novel method on edge that aims to evaluate an in-home exercise program in this paper. Our key idea is to integrate multiple different vision-based pre-trained models. In the proposed method, we mainly integrate and call multiple pre-trained models to recognize human facial, skeletal, and hand movements with edge computing. Also, we propose to evaluate an in-home exercise program process from the Euclidean distance, velocity, and angle of the two-dimensional coordinates of feature points in the time series. Through experiments, we estimate the physical health status from the analysis of body movements in the elderly.