Marc Marais

GOLF SWING SEQUENCING USING COMPUTER VISION

Supervisor: Dr Dane Brown

Analysis of golf swing events is a valuable tool to aid all golfers in improving their swing. Image processing and machine learning enable an automated system to perform golf swing sequencing using images. The majority of swing sequencing systems involve the use of expensive camera equipment or a motion capture suit. Therefore, an image-based swing classification system is proposed. It is evaluated on the GolfDB dataset. The system implements an automated golfer detector and traditional machine learning algorithms to classify swing events.

Overall, the best performing classifier, the LinearSVM, achieved an recall score of 88.25%  on the entire GolfDB dataset. The system outperformed McNally et al. (2019) Bi-LSTM deep learning approach to achieve swing sequencing, which achieved a percentage of correct events of 76.1% on the same GolfDB dataset. Overall, the results were promising and work towards a system that can be implemented to assist all golfers in swing sequencing without the need for expensive equipment.

A high-level overview of the proposed system can be found below:

References

William McNally, Kanav Vats, Tyler  Pinto,  Chris  Dulhanty,  John  McPhee, and Alexander Wong. GolfDB: A video database for golf swing sequencing. IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, 2019-June:2553–2562, 2019. ISSN 21607516. doi: 10.1109/CVPRW.2019.00311.

Please see the Kaggle_repository_readme file for links to the code and dataset files.