In this work, an analysis of short-term and long-term trackers performance on marine-based dataset of video sequences is presented. To this end we used already existing marine datasets to create two new marine datasets. The first dataset is suitable for short-term tracking analysis, while the other one is suitable for long-term tracking analysis. We perform experiments on those datasets using some of the top-performing short-term and long-term trackers based on their Visual Object Tracking competitions performance. The results of short-term tracking are presented in form of expected average overlap, average overlap, number of failures and expected average overlap curves. Processing speed of short-term trackers is also evaluated. In addition to the supervised experiment, realtime experiment is performed as well. The results for long-term trackers are presented as a precision-recall graph and F-score graph. Long-term trackers are evaluated based on their processing speed as well. The results are analyzed and compared to results in similar studies.
|
