Evaluation of Bounding Box Level Fusion of Single Target Video Object Trackers

摘要

The main objective of this work is to evaluate a simple fusion system which improves the performance of several object trackers, within a methodological and rigorous evaluation framework. The considered algorithms are monocamera single target trackers. After analyzing in detail the state of the art, an evaluation framework is selected and presented. The sequences selected in this evaluation try to represent different real scenes and conditions. Then, clasical and modern tracking algorithms are selected and evaluated individually, in order to understand their performance in different scenarios and problems. Finally, some fusion methods are described and evaluated, comparing their results with the results of the individual tracking algorithms.