Exercise instruction support application using 3D pose estimation with monocular camera

Abstract

This research is aimed at developing a simple support application that allows general users to easily approximate the ideal motion. The application performs 3D pose estimation using “Lifting from the Deep” from monocular camera images that can be easily acquired by the user, and evaluates whether the target motion is close to the ideal teacher motion. The evaluation is performed based on the DTW distance of the difference data of the target time series parameter. The proposed application has been shown to be useful for supporting exercise guidance in rowing experiments.