Fig 1.
Steps for the construction of images of the convex polygons that represent the organization of the teams in the field: (A) Position of the players of each team at a given time; (B) Application of the quickhull technique for the construction of the convex hull; (C) Convex hull polygon in the respective field position; Images of the shapes analysed for each team, (D) Team 1 and (E) Team 2.
Fig 2.
(A) Image of a polygon; (B) dilated contours; (C) Logarithmic function of the areas of exact dilation radius; (D) Multiscale Fractal Dimension of the analysed image, with Maximum Fractal Value (MF), Multiscale Fractal Curve (MFC), Fractal area (FA).
Table 1.
Results of the interpersonal agreement analysis to identify the relevant images for each descriptor among the 3 different evaluators (A, B, and C).
Table 2.
Results of the percentage of agreement between images considered not relevant (NR—NR) and images considered relevant (R—R) between 3 different evaluators.
Fig 3.
Representation of retrieved performance results in relation to the precision x recall curve for all evaluated descriptors.
Abbreviations: FA = Fractal Area, MF = Maximum Fractal, MFC = Multiscale Fractal Curve.
Table 3.
Values of mean ± standard deviation for precision (P@5 and P@10) and recall (R@5 and R@10) for evaluating the efficacy of different descriptors.
Fig 4.
Result of clustering of MFC for Teams 1 and 2: (A) and (B), 1° and 2° half of Team 1, respectively; (C) and (D), 1° and 2° half of Team 2, respectively. Results too of the shapes found for each analysed cluster, when they occur throughout the match (bar graphic) and percentiles of shapes in each cluster (pie graphic).