Golf Shot Swing Recognition Using Dense Optical Flow

This paper introduces a computer vision-based methodology for identifying golf swings. The proposed approach utilizes videographic data of golf swings and recognizes swing type and postures such as Backswing, Downswing, and Follow Through. The system experiments with techniques such as Dense and Sparse Optical Flow to identify the Golf club's trajectory. The features of these trajectories are then extracted using Histogram of Oriented Gradients. To create a resilient model, the feature of golfer's posture during a specific swing is extracted using Scale-Invariant Feature Transform with morphological image frame. The extracted features are normalized, concatenated, and classified using various classification techniques. When evaluated on the standard GolfDB dataset, the system achieves an Accuracy of 84.4 percent for Random Forest classification with a Precision Score of 0.7949. This study proposes a promising approach to identify golf swings using computer vision techniques, which could potentially have implications in enhancing performance analysis in the field of sports.