Rugby Goal Kick Prediction Using OpenPose Coordinates and LSTM

The goal kick is the most delicate play in rugby. To train athletes, giving accurate instructions is difficult because each player’s form is different. Furthermore, in Japan, rugby goal posts are only installed in a few areas, so athletes who do not have access to a goal post have limited possibilities to perform practical kicks, and goal kick practice is insufficient. As a result, we used Long Short-Term Memory (LSTM) to create a goal prediction model from goal-kick videos to provide the player with feedback. We also attempted to determine which parts of the body are crucial criteria for scoring. This paper addresses only goal kicks/conversion kicks and penalty goal kicks made from stationary locations, not kicks made while the ball was in play such as the drop-kick. According to the findings, the model built using domain expertise was just as precise as the model built using all joint data. This result proved that the right knee and ankle of the kicking leg, as well as the positions of the right eye and shoulder, are crucial elements in determining a successful kick.