Scholarlink

<p>The SiamRPN object tracking algorithm achieves a good balance between speed and performance but struggles with tracking accuracy, especially in complex backgrounds. This study introduces an inverted residual mobile block to improve the algorithm's ability to distinguish interferences and objects, thereby enhancing tracking accuracy and robustness. By incorporating multi-head self-attention and depthwise separable convolutions, the proposed block decomposes and recombines feature information from the Siamese network backbone, effectively avoiding incorrect tracking caused by similar objects. Extensive experiments on OTB2015, UAV123, and VOT2016 datasets demonstrate that our method significantly outperforms the baseline SiamRPN algorithm in terms of tracking accuracy and robustness, while maintaining a high tracking speed. Notably, on the UAV123 dataset, our approach exhibits excellent performance, indicating its potential application in unmanned aerial vehicle tracking and enemy search. The code is available at <a href="https://github.com/jasonzhengye/ISiamDW/tree/master">https://github.com/jasonzhengye/ISiamDW/tree/master</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced visual tracking with inverted residual mobile block: a SiamRPN-based approach

  • 期刊论文

  • 发表日期：

    2025年5月28日

查看全文
  • Zhengye Jiang,
  • Tianping Li

摘要

The SiamRPN object tracking algorithm achieves a good balance between speed and performance but struggles with tracking accuracy, especially in complex backgrounds. This study introduces an inverted residual mobile block to improve the algorithm's ability to distinguish interferences and objects, thereby enhancing tracking accuracy and robustness. By incorporating multi-head self-attention and depthwise separable convolutions, the proposed block decomposes and recombines feature information from the Siamese network backbone, effectively avoiding incorrect tracking caused by similar objects. Extensive experiments on OTB2015, UAV123, and VOT2016 datasets demonstrate that our method significantly outperforms the baseline SiamRPN algorithm in terms of tracking accuracy and robustness, while maintaining a high tracking speed. Notably, on the UAV123 dataset, our approach exhibits excellent performance, indicating its potential application in unmanned aerial vehicle tracking and enemy search. The code is available at https://github.com/jasonzhengye/ISiamDW/tree/master.