Brouwer, M.J. (Max) (2021) Classification of tennis serve direction using supervised machine learning. thesis, Sport Sciences.
Full text available on request.Abstract
Tennis match statistics are often derived from competition footage via manual annotation. However currently, manual annotation is a time consuming, inefficient task. Recent efforts to (semi-)automate the detection of tennis rally events have not succeeded in showing satisfactory reliability yet, and/or are not available in single-camera set-ups. Recent technological innovations like automated player position tracking from video footage can provide data that could facilitate in speeding up tennis match annotation. Additionally, supervised machine learning models allow for scientists to explore scientific issues from a new perspective. In this study, we therefore explored a new method to classify tennis serve direction, adopting a supervised machine learning approach. Using the pose and position estimator wrnchAI, player tracking data of 1841 rallies from nine 2019 ATP matches were extracted from birds-eye video footage. Features (e.g. machine learning input variables) were constructed with tracking data from the server and returner. An expert observer classified the serve direction of all rallies in one of three position-based classes, adhering to predefined border definitions. Three machine learning models were trained, tested and extensively evaluated on relevant metrices. The random forest model achieved an accuracy of 80% for classifying serve direction, thereby outperforming the other models. When implementing this model, the annotation of tennis rally events, like serve direction, can be sped up drastically. Additionally, by adhering to a single-camera set-up, the proposed method is applicable and accessible. Keywords: machine learning, spatiotemporal analysis, tennis, annotation
| Item Type: | Thesis (UNSPECIFIED) |
|---|---|
| Supervisor name: | Poel, dr. H. de and Kempe, dr. M. |
| Faculty: | Medical Sciences |
| Date Deposited: | 11 May 2022 10:27 |
| Last Modified: | 11 May 2022 10:27 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/3289 |
Actions (login required)
 |
View Item |