Visual Object Tracking Using Deep Learning Techniques: A Comparison
Enzheng Su · Transactions on Computer Science and Intelligent Systems Research · 2024
Visual object tracking is one of the important topics in computer vision: given a target object in the first frame of a video, follow this target object on subsequent frames. Its applications are found in autonomous driving, human-computer interaction, military operations, and game development. However interesting these applications may be, the implementation remains very challenging because occlusions happen among other factors like scale variations or rotations that require a tracker to be robust and generalize well to give accurate results. In this paper, we provide an extensive review of visual object tracking methods divided into three primary frameworks: correlation filter-based trackers, Siamese network-based trackers, and Transformer-based trackers. We provide insights into a total of 12 state-of-the-art techniques for each category through large-scale experimentation on various benchmark data sets, aiming to reveal the research frontier of the visual object tracking field and point out some potentially useful directions for future investigations.