Occlusion Detection for Long Term Visual Tracking
Zhongli Wang, Baigen Cai, Yingshuang Mei, Yifu Wang, G.Y. Lv, Ruizeng Shan, M. Zhang · 2018
How to detect the occlusion and design a tracker which is robust to occlusion are very difficult in the field of visual tracking. To get a comprehensive understanding about this issue, the trackers or methods which are like to be robust to the occlusion are briefly summarized. Generally, the metrics of occlusion is dependent on the representation of target appearance model. It can be coarsely classified into three types according to the existing literatures, 1) Part-based or structure-based representation; 2) Key-point based representation; 3) Segmentation-based method. The working principles of these three models are introduced. To get a deep insight about the performance of them, evaluation experiments on the occlusion metric are conducted. Finally, some issues for long-term visual tracking are discussed.