Correlation study of video object trackers evaluation metrics
A correlation study of a set of video object trackers evaluation metrics is presented. There are multiple metrics in the state-of-the-art, and the main differences between them are based on the penalties that are attributed to the errors (false positives, false negatives, target loss etc.). After defining each one of the considered metrics, a correlation study is presented showing the redundancy between all the metrics. This reported study was performed using multiple tracking algorithms, and an extensive set of video sequences that attempt to cover many different situations.