Answers are grounded in this paper and cite their sources.
Background Markerless motion capture can provide a novel method for workload monitoring in tennis, with mechanical work (external and internal components) previously presented as a suitable metric. However, markerless methods typically involve large amounts of raw data and long data processing times, hence this is currently an impractical solution. Therefore, this study investigated the effects of reducing system specifications, with the aim of finding a compromise between computational cost and accuracy that is suitable for use in applied settings. Methods Video data of 15 tennis players completing an on-court fatiguing protocol were processed with a markerless motion analysis pipeline (based on open-source pose estimation) to calculate mechanical work. Eight cameras operating at a frame rate of 200 Hz and 1,920 × 1,080 resolution were used. Data were reprocessed using fewer cameras, lower frame rates and lower resolutions. Results Reducing the number of cameras, even by one, or resolution resulted in large drop-offs in performance, related to erroneous initial pose estimation, with root mean squared differences (RMSD) of ∼160% (external) and ∼150% (internal) when using four cameras and ∼640% (external) and ∼190% (internal) when using quarter resolution videos. Conversely, frame rates of 50 and 100 Hz were deemed acceptable for calculating external (RMSD: 1.4%) and internal (RMSD: 0.9%) work respectively. Reducing the frame rate was also associated with the largest reduction in computational cost, with data saving and processing times approximately proportional to the number of frames. Conclusion Within the conditions examined, reducing the number of cameras or resolution was not an option due to inaccuracies of the open-source pose estimation algorithm, but utilising a lower frame rate appears to offer a promising compromise between computational efficiency and measurement accuracy, potentially increasing the viability of markerless motion capture as a solution for in-field workload monitoring.
No takes yet. Share an insight, caveat, or question.
Emmerson et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: