A Novel Deep Learning Model for Automated Golf Swing Pattern Recognition in Video Footage

The automated classification of golf swing events is a crucial procedure for analyzing swings using video analysis. However, there has been a lack of initiative in applying machine learning (ML) algorithms to accurately determine and classify the various stages of a golf swing. This paper proposes a Customized Convolutional Neural Network (CNN) to classify the golf swing phases into eight primary phases, encompassing the pre and post-swing events, based on open source Kaggle dataset. The frames from the videos are extracted and converted into structured data along with the player’s data details. The combined data is pre-processed using Linear discriminant analysis. The proposed customized CNN model is compared with the ML models like Support Vector Machines and the CatBoost. The proposed CNN model outperforms by achieving an accuracy of 84.60%. Furthermore, the proposed approaches can separate all the swing phases only based on video data. This suggests that machine learning-based swing classification techniques may be utilized in different motion-analysis settings by splitting motion phases with more flexibility in video analytics in real time.