Robust Object Tracking via Hierarchical Estimation in SiamFC Network

Yueyang Gu, Yu Qiao, Kuan Xu, Jie Yang, Jiaxuan Xie, Lu Wang, Mingna Liu
2019 Australian Journal of Intelligent Information Processing Systems  
Recently, SiamFC network has been investigated into object tracking community. This network utilizes two branches sharing same parameters as feature extractor and cross-correlation as similarity metric. However, most of SiamFC trackers discard low level conv-feature and only take the conv-feature from last layer into account. In this paper, we propose a hierarchical estimation framework for SiamFC network to take full advantage of both low level and high level conv-features. We firstly
more » ... hierarchical SiamFC networks using pre-trained VGGNet, ResNet and DenseNet. While tracking, hierarchical SiamFC network can infer target position via different conv-features. We also define adaptive weights for each score map from different layers. The final position of target is estimated by score map with maximum weighted PSR. Experimental results demonstrate that SiamFC trackers equipped our proposed framework outperform baselines and achieve state-of-art performance.
dblp:journals/ajiips/GuQXYXWL19 fatcat:q5c4rfefizbhfpipyazb22vbqe