作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2022, Vol. 48 ›› Issue (1): 288-295,304. doi: 10.19678/j.issn.1000-3428.0060045

• 开发研究与工程应用 • 上一篇    下一篇

基于孪生区域建议网络的无人机目标跟踪算法

杨帅东, 谌海云, 徐钒诚, 赵书朵, 袁杰敏   

  1. 西南石油大学 电气信息学院, 成都 610500
  • 收稿日期:2020-11-18 修回日期:2020-12-21 发布日期:2022-01-04
  • 作者简介:杨帅东(1998-),男,硕士研究生,主研方向为图像处理、视觉目标跟踪;谌海云,教授;徐钒诚,硕士研究生;赵书朵,副教授、硕士;袁杰敏,助理实验师、硕士。
  • 基金资助:
    南充市科技项目(19SXHZ0019,19SXHZ0011)。

Target Tracking Algorithm Based on Siamese Region Proposal Network for UAV

YANG Shuaidong, CHEN Haiyun, XU Fancheng, ZHAO Shuduo, YUAN Jiemin   

  1. School of Electrical Engineering and Information, Southwest Petroleum University, Chengdu 610500, China
  • Received:2020-11-18 Revised:2020-12-21 Published:2022-01-04

摘要: 在无人机跟踪过程中,遮挡、光照变化、背景干扰等影响会导致跟踪目标丢失。基于SiamRPN算法提出一种无人机目标跟踪算法。通过在网络中加入空间条带池和全局上下文模块建立远程上下文关系,以适应不同的跟踪场景。同时利用改进交并比的计算方法提取目标特征,并回归精准的预测框。在UAV123数据集上的实验结果表明,相比SiamRPN、SiamFC、SAMF等算法,该算法的跟踪性能较优且具有较强的鲁棒性,尤其在背景干扰环境下,其精确率和成功率较SiamRPN算法分别提升了6.54%和11.63%。

关键词: 目标跟踪, 无人机, 深度学习, 孪生网络, 注意力机制

Abstract: When performing target tracking tasks, Unmanned Aerial Vehicles(UAV) often lose their targets due to occlusion, illumination changes and background interference.Based on the SiamRPN algorithm, a target tracking algorithm is proposed for UAV.The algorithm first adds a strip pooling and a global context module to the network, which establishs remote context and enables it to adapt to different tracking scenarios.Then the calculation method based on cross-comparison optimization is used to complete the feature extraction of the target, and to return an accurate prediction boundary frame.The experimental results on the UAV123 public data set benchmark show that compared to SiamRPN, SiamFC, SAMF algorithms, the algorithm exhibits better tracking performance and high robustness.Compared to SiamRPN algorithm, the accuracy rate and success rate of the algorithm are improved by 6.45% and 11.63% respectively in the background interference.

Key words: target tracking, Unmanned Aerial Vehicle(UAV), Deep Learning(DL), siamese network, attention mechanism

中图分类号: