Visual Object Tracking: Method and Comparison
Qiangyu Li, Letian Quan, Zhuangzhi Sun, Yuchen Wu · 2022 IEEE 2nd International Conference on Electronic Technology, Communication and Information (ICETCI) · 2022
Visual Tracking is an important branch of computer vision, it begins with a frame, identify target objects from different backgrounds, modeling target's appearance, analysis and modeling target's motion information in the subsequent frames, so as to continuously predict the motion state of the target, and determine its position in each frame of image, and finally realize the tracking of an object. In recent years, many excellent and classic algorithms and frameworks have emerged in this field, this is divided into two main categories, which are correlation filtering methods and deep learning methods. They have their own advantages. For example, excellent correlation filters can achieve high-speed and stable tracking effects. With the help of deep learning networks, target features can be obtained in advance to achieve more accurate tracking. In this paper, we have summarized 15 papers covering four directions or fields of Correlation Filter, Siamese Network, Multi-domain Network and Transformer. The summary includes 2 frameworks and 13 algorithms or methods. We also analyzed the advantages and disadvantages of the summarized algorithms, and gave a performance comparison between different algorithms.