HUDSON, Christopher (2015). Automated tracking of swimmers in the clean swimming phase of a race. Doctoral, Sheffield Hallam University.
Hudsonautomatedtracking.pdf - Accepted Version
Available under License All rights reserved.
Download (6MB) | Preview
The current advice for a sports analyst when filming a large performance area is to use multiple fixed cameras or a single panning one. Neither of these options is ideal: multiple cameras must be positioned, have their shutters synchronised and their footage combined for analysis; a panning camera makes it difficult to determine an athlete’s movement relative to an external frame of reference. The aim of this study was to establish a process that enabled the confident, accurate and precise use of a wide field of view for measuring distance and speed in large performance areas. Swimming was used as an example sport as it had a large performance area, which measured 50 m by 25 m. A process for determining the accuracy and precision with which distance and speed could be reconstructed from a wide field of view was developed. A nonlinear calibration procedure was used to account for radial distortion. The Root Mean Square Error (RMSE) of reconstructed distances for a wide field of view was 16 x 10-3 m. This compared favourably with a three camera system reported in the literature, which had an RMSE of 46 x 10-3 m. In addition, it was shown that a wide field of view could be used to identify a 1% enhancement in speed when it was measured over 10 m or more. A wide field of view was used to capture video footage of a swimming competition. This was used to track swimmers using two methods: manual and automated. The two methods showed good agreement for mean speed, but the automated one had higher variability in instantaneous speed than did the manual.
|Item Type:||Thesis (Doctoral)|
|Research Institute, Centre or Group:||Centre for Sports Engineering Research|
|Depositing User:||Helen Garner|
|Date Deposited:||24 Sep 2015 11:09|
|Last Modified:||24 Sep 2015 16:30|
Actions (login required)
Downloads per month over past year