金久才, 张家林, 刘德庆, 等, xxxx. 基于改进SiamRPN的高速无人船海上目标视觉跟踪研究[J]. 海洋科学进展, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230629001.
引用本文: 金久才, 张家林, 刘德庆, 等, xxxx. 基于改进SiamRPN的高速无人船海上目标视觉跟踪研究[J]. 海洋科学进展, x(x): xx-xx. doi: 10.12362/j.issn.1671-6647.20230629001.
shu
JIN J C, ZHANG J L, LIU D Q, et al, xxxx. Visual tracking of maritime targets for a high-speed unmanned surface vehicle based on improved Siamese region proposal network[J]. Advances in Marine Science, x(x): xx-xx. DOI: 10.12362/j.issn.1671-6647.20230629001
Citation: JIN J C, ZHANG J L, LIU D Q, et al, xxxx. Visual tracking of maritime targets for a high-speed unmanned surface vehicle based on improved Siamese region proposal network[J]. Advances in Marine Science, x(x): xx-xx. DOI: 10.12362/j.issn.1671-6647.20230629001
shu

基于改进SiamRPN的高速无人船海上目标视觉跟踪研究

Visual Tracking of Maritime Targets for a High-speed Unmanned Surface Vehicle Based on Improved Siamese Region Proposal Network

    摘要
  • 摘要: 对于海上目标的监视侦察，无人船具备成本低、危险性小和隐蔽性好等优势。针对船只目标的抵近监视与跟踪问题，基于自研发的“久航750”7 m级高速无人船目标监视系统，本文提出了一种基于改进孪生候选区域生成网络（Siamese Region Proposal Network, SiamRPN）的海上目标视觉跟踪方法。该方法以ResNet50网络为主干网络，并在Siamese和RPN子网络之间增加了多层特征融合模块，通过对比原始SiamRPN算法、SiamFC（全卷积孪生网络）算法，显示了算法的鲁棒性和准确性。利用该高速无人船抵近监视“智飞号”无人集装箱船的海上数据，完成了由远及近过程中的无人船低速、加速和高速情况下的视觉跟踪测试，试验结果验证了所提算法的可行性和有效性。

     

    Abstract: Unmanned Surface Vehicle (USV) has many advantages for close in monitoring and tracking of ships, such as low cost, safe and sneak. A visual tracking method for a target monitoring system based on 7 m-level ‘Jiuhang 750’ high-speed USV is proposed using improved Siamese Region Proposal Network (SiameseRPN). The backbone network of the proposed algorithm is replaced by ResNet50, and multi-level feature fusion modules are added between the Siamese network and RPN network. Comparing with the original SiamRPN and SiamFC, the accuracy and robustness of the proposed algorithm is improved. Based on the monitoring data of ‘Zhifei’ container using the ‘Jiuhang 750’ USV in the sea experiment, the visual target tracking algorithm is tested and validated under three conditions, which are unstable course at low-speed, speeding and high-speed conditions. The feasibility and availability of the proposed algorithm are verified by the experiment results.

     

    • HTML全文
    • 参考文献(10)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回