Baseball Swing Pose Estimation Using OpenPose
Yung-Che Li, Ching-Tang Chang, Chin-Chang Cheng, Yu-Len Huang
Abstract
This study explores usefulness of using the human pose estimation technique in sport. Since the expansion of deep learning techniques, human pose estimation became an important field of computer vision, it can be used in many applications like pose analysis, correction, training session, etc. The proposed method is used to estimate whether a baseball hitter performs a good swing. The hitter's limb coordinates are detected by the OpenPose model which is a real time multi-person detection system. The coordinates are used to calculate hip distance and limb angles, then the distance and angles are applied with our custom rules. The custom rules are made by researches and coaching experience in order to evaluate the swing of baseball hitter. Each rule is awarded differ points by its importance which we assumed. The goal of this study is using technology assistance in sport coaching scenario.