On-Court Workload Assessment of Tennis Players Using Markerless Motion Analysis

: (Alternative Format Thesis)

Abstract

Workload monitoring is important in tennis for optimising performance and reducing injury risk, but effective strategies are not well-established. The emergence of deep learning-based markerless motion capture technologies opens up possibilities for non-invasive measurement of full-body kinematics in the field, with potential applications in workload monitoring. The purpose of this thesis was to develop a video-based markerless motion analysis tool for the application of workload monitoring in tennis, with a particular focus on mechanical work as a potential metric. The first study evaluated the agreement between a bespoke markerless motion capture system, utilising open-source pose estimation, and de facto marker-based motion capture and ground reaction force methods for calculating mechanical work during tennis-specific movements. Results suggested good agreement between all methods, albeit with a systematic underestimation by the markerless method as well as additional noise in the energy timeseries due to high-frequency artefacts in the markerless kinematics. The agreement was deemed acceptable for workload monitoring applications. The second study explored the use of the markerless system for the application of workload monitoring in a real-world environment, i.e. a tennis court. The association between mechanical work performed during a fatiguing protocol and a measure of acute neuromuscular fatigue was assessed. Strong correlations were found, suggesting that this approach can provide an indication of workload. Moreover, simpler estimations of mechanical work also yielded strong correlations, despite inaccurate absolute values, indicating that these may be suitable for some applications. The third study investigated the effects of reducing specifications of the markerless system, with the aim of finding a compromise between computational cost and accuracy that would be suitable for use in applied settings. Results indicated that the number of cameras and resolution should be kept as high as possible to minimise errors related to pose estimation inaccuracies, but lowering the frame rate to 50 Hz could be an effective way to reduce the computational cost whilst maintaining sufficient accuracy. This thesis has demonstrated that mechanical work can be accurately measured with markerless motion capture and that it can be used in an ecologically-valid setting to indicate player workload, whilst utilising a lower frame rate makes this approach more viable for in-field workload monitoring. These findings highlight the future scope for the implementation of this approach as a non-invasive on-court workload monitoring strategy.

Date of Award	25 Jun 2025
Original language	English
Awarding Institution	University of Bath
Supervisor	Steffi Colyer (Supervisor), Laurie Needham (Supervisor) & Sean Williams (Supervisor)