We attempted to find a subset model that would allow robust prediction of a swimmer's vertical body position during front crawl with fewer markers, which can reduce extra drag and time-consuming measurements. Thirteen male swimmers performed a 15-metre front crawl either with three different lung-volume levels or various speeds, or both, without taking a breath with 36 reflective markers. The vertical positions of the centre of mass (CoM) and four representative landmarks in the trunk segment over a stroke cycle were calculated using an underwater motion-capture system. We obtained 212 stroke cycles across trials and analysed the vertical position derived from 15 patterns as candidates for the subset models. Unconstrained optimisation minimises the root-mean-square error between the vertical CoM position and each subset model. The performance evaluated from the intra-class correlation coefficient (ICC) and weight parameters of each subset model were detected from the mean values across five-fold cross-validation. The subset model with four markers attached to the trunk segment showed good reliability (ICC: 0.776 ± 0.019). This result indicates that the subset model with few markers can robustly predict a male swimmer's vertical CoM position during front crawl under a wide range of speeds from 0.66 to 1.66 m · s-1.
Keywords: Underwater motion-capture system; body position; intra-class correlation coefficient; simple model; swimming.