Fig 1.
A) Calibration tools. B) Action sport cameras (GoPro, Hero3+, Black). C) Rigid bar used to 3D reconstruction accuracy evaluation. D) Camera position and acquisition volume.
Fig 2.
Schematic workflow of the two-stage camera calibration for a generic number of cameras.
Fig 3.
Three different calibration wand movements to evaluate the calibration dependability.
Table 1.
Results of the 5 trials of dynamic rigid bar test (HIGHRES and LOWRES).
Nominal distance dn between the two markers: 250mm.
Fig 4.
The histograms of the residual error distribution (cumulated over the five trials) for HIGHRES (1920–1080) and LOWRES (1280–720).
The average values were 1.28 and 2.41mm, respectively.
Table 2.
Minimum, mean and maximum inter-marker distance, averaged across 5 trials and the corresponding mean absolute error (dn: 250mm).
M1-1 (one marker with zig-zig movement); M2-1 (one marker with circular movement); M3-1 (one marker with up-down); M1-2 (two markers with zig-zig movement); M2-2 (two markers with circular movement). The post-hoc comparison results were reported (*p<0.05).
Fig 5.
Two instants of the front-crawl swim cycle surveyed by two cameras.
The swimmer is equipped with surface markers attached to the right arm. This real condition elucidates how poor image contrast and water disturbance can complicate the automatic marker detection on the image and tracking analysis, affecting the accuracy of the 3D kinematic analysis.