WCSE

Research on Target Tracking Algorithm Combining Hierarchical Features and Hybrid Attention

Abstract— In tasks of target tracking, Fully-Convolutional Siamese networks for object tracking (SiamFC) has poor tracking performance in complex scenarios such as illumination effects, scale changes, and target deformation. To this end, a target tracking algorithm combining hierarchical features and hybrid attention is proposed. First, the improved VGG19bn is used as the backbone network to extract more adequate target features; Secondly, hybrid attention is embedded at the end of the template branch of the Siamese network to make the target information more prominent; Meanwhile, add online update mechanism for target template to improve the long-term tracking ability of the model; Finally, the filtered low-level response map and the high-level response map are adaptively weighted and fused to improve the network's ability to distinguish positive and negative samples. Tested on the OTB2013 dataset, the accuracy and success rate of the proposed algorithm are 5.3% and 5.2% ahead of the SiamFC algorithm; Tested on the OTB2015 dataset, the accuracy and success rate of the proposed algorithm are 8.4% and 6.7% ahead of the SiamFC algorithm. The experimental results show that the proposed algorithm has great advantages in tracking accuracy, and has better robustness when dealing with complex scenes.

Index Terms— Object tracking; Siamese network; Attention mechanism; Online update mechanism; Hierarchical feature fusion

Caixia Meng
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications; Shannxi Key Laboratory of Network Data Analysis and Intelligent Processing, China; Xi’an Key Laboratory of Big Data and Intelligent Computing, China
Kaijie Xi
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications
Bing Zhao
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications
Hongpeng Chu
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications
Jiabao Zhang
School of Computer Science and Technology, Xi'an University of Posts and Telecommunications

[Download]

Cite: Caixia Meng, Kaijie Xi, Bing Zhao, Hongpeng Chu, Jiabao Zhang, "Research on Target Tracking Algorithm Combining Hierarchical Features and Hybrid Attention," WCSE 2022 Spring Event: 2022 9th International Conference on Industrial Engineering and Applications, pp. 61-71, Sanya, China, April 15-18, 2022.