Figure 4
Ranking of the automatic trackers and the manual operators regarding the performance metrics precision (a), processing speed (b) and spatial robustness (c,d). (a) The precision at 5 mm tracking error compared to the randomized operator data (OP1), as indicated by the dashed line in Fig. 3a. Out of all automatic trackers, KCF performs best. (b) The trackers regarding their frame rate for an average tracking iteration. For evaluation of the spatial robustness, automatic tracking was repeated 125 times for each dataset with initial bounding boxes varying in size and position. The boxplots in (c) show the mean and in (d) the standard deviation of these sets of 125 precision curves at 5 mm tracking error. Regarding spatial robustness, the KCF tracker is clearly favorable with highest average precision and very low standard deviation. More detailed explanations on the spatial robustness evaluation can be found in the Supplementary Materials Suppl. Fig. S1a.
