Recurrent Neural Network for Evaluating Rhythmic Movement during Radio Calisthenics Using A Markerless Motion Capture System Implemented on Smart Devices

Japan is currently the challenge of a rapidly aging population, which has increased the need for preventive healthcare. However, there is a shortage of caregivers who can provide appropriate advice to elderly individuals. Previous researches have focused on developing a health promotion system that can assess proper exercise movements and provide improvement advice. However, most of these studies use joint angles as indicators to evaluate exercise movements, and few have focused on evaluating the rhythm of exercise movements. Rhythm can be an indicator of declining physical function in the elderly and can also provide users of the system with the enjoyment of moving their bodies. In this paper, we propose a method for evaluating the rhythm of exercise movements using a smart device. In this system, a recurrent neural network is used for posture recognition, and the obtained output waveform is evaluated for rhythm by applying the autocorrelation function.