Deep Learning-Based Prediction of Football Players’ Performance During Penalty Shootout

The goal of this study is to analyse football players’ body positioning data during the penalty to predict the region of the goalpost where kickers will shoot. YOLOv4 was initially used on a custom-created video dataset to detect the goalkeeper, kicker, football, and goalpost. Following that, OpenCV was used to track the football and divide the goalpost into four zones. To track the kicker’s body positioning data, pose estimation was used. After recording the data for posture estimation, an LSTM model was utilized to recognise a footballer’s activity. The dataset contains 1560 numpy files with a total of 205920 pose landmarks from pose estimation. The LSTM model attained an accuracy of around 50 percent on the test dataset. Before 15 seconds, 10 seconds, 5 seconds and 1 second of the 20 penalty shoot clips, this study achieved mean accuracy of 9.6 percent, 26.2 percent, 52.80 percent and 79.05 percent, respectively.