Golf swing biomechanics for virtual sports training applications

Abstract

Sports biomechanics is a highly versatile area of research that leverages multidisciplinary skillsets to enhance game performance and reduce the risk of injury for athletes. By employing techniques such as inverse dynamics, kinetics, and kinematics, numerous challenges in sports performance can be addressed. Machine learning algorithms and Mathematical modeling can now provide feasibly innovative and effective solutions. This study focuses on the sport of golf, emphasizing its importance for amateur golfers. The research is important because, for amateur golfers, not only golf swing but finding the right balance between club speed, club selection, and spin can greatly impact their performance on the course. Experimenting with different clubs and understanding how club speed influences spin can help golfers make more informed decisions to improve their game. Two important research questions are addressed in this work, (i) ball detection; (ii) calculation of accurate carry distance calculation and magnus effect. Golfer’s swing is envisioned in the future extension of this research incorporating golfer’s biomechanics assessment as well as postural stresses. Advanced computer vision and object detection algorithm were employed specifically utilizing YOLOv5, to analyze golf swings captured on smartphones. By accurately measuring critical parameters such as initial velocity, angle, and spin rate, and integrating these with a mathematical model it was possible to calculate the golf ball’s trajectory. Yolov5, a convolutional neural network model, was trained on a dataset comprising approximately 24000 images out of which 18,000 images were selected for 300 epochs after dataset cleaning. By measuring key parameters such as xiii initial velocity, angle, and spin rate, and integrating these data with a mathematical model, the golf ball’s trajectory was calculated. YOLOv5, a convolutional neural network model, was trained on a dataset of approximately 24,000 images, with 18,000 images selected for training over 300 epochs after dataset cleaning to remove improperly labeled images. The training utilized the YOLOv5s.yaml configuration on a Single GPU Asus Rog Strix Nvidia RTX 3060 TI with 32 GB of DDR4 CL 18 RAM, incurring a total computational cost of 11 hours for this specific use case. To predict the trajectory of the golf ball, specific differential equations derived from Newton’s Second Law of Motion, incorporating air resistance (drag) and the Magnus effect (lift), were used. These equations were solved numerically using Euler’s method, allowing for precise modeling of the ball’s flight path. Results illustrate that YOLOv5 accurately detects golf balls in various lighting and background conditions. The mathematical model reliably predicts the ball's trajectory and range, aligning closely with actual measured outcomes. This demonstrates the practicality and effectiveness of smartphone-based sports analysis backed up Machine learning algorithm and mathematical modeling.