Object Tracking Based on Adaptive Multi-Template Fusing
Ziliang Guo, Zihao Li, Xiwen Zhang, Xingqi Fang, Yi Tian, Xiaowei Zhang, Yu Qiao · 2024
Target lost and robustness to occlusion scenarios hinder real-world applications of current object tracking methods. In this paper, we focus on reducing the frames of target lost in the single object tracking task. We propose a multi-template object tracking framework which incorporates occlusion detection. Firstly, the multi-template design memorizes long-term target appearance, and adaptively fuses into a unified template based on short-term similarity knowledge. Secondly, the evaluated occlusion severity guides the template updating. As a result, the insensitivity to the learning rate is achieved on VOT2016 and VOT2018. The significant decrease in frames of lost target further highlights the superiority of our approach.