改进野狗算法优化的车辆跟踪框架
CSTR:
作者:
作者单位:

(1.长安大学 电子与控制工程学院,西安 710064;2.西安市智慧高速公路信息融合与控制重点实验室(长安大学), 西安 710064;3.长安大学 信息工程学院,西安 710064)

作者简介:

王志伟(2001—),男,硕士研究生;黄鹤(1979—),男,教授,博士生导师

通讯作者:

黄鹤,huanghe@chd.edu.cn

中图分类号:

TP391

基金项目:

国家自然科学基金(52472446);陕西省留学人员科技活动择优资助项目(2023001);中央高校基本科研业务费资助项目(300102325501);中国交通教育研究会教育科研课题(JT2024YB444)


Improved vehicle tracking framework optimized by dingo algorithm
Author:
Affiliation:

(1.School of Electronic and Control Engineering, Changan University, Xian 710064, China; 2.Key Laboratory of Intelligent Expressway Information Fusion and Control(Changan University), Xian 710064, China; 3.School of Information Engineering, Changan University, Xian 710064, China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    为提高道路监控中车辆目标跟踪的准确性,特别是面向目标车辆由于距离变化和遮挡导致的跟踪精度降低以致跟踪对象丢失问题,提出一种基于改进野狗优化算法的车辆目标跟踪框架(IDOAT)。首先,该框架融合快速方向梯度直方图特征和深度特征,以增强对目标车辆的表征能力;其次,结合卡尔曼滤波初始化和螺旋更新机制,对野狗优化算法进行改进,以适应目标在搜索过程中容易丢失的问题;最后,采用改进的野狗优化算法作为搜索策略,快速定位目标,并选取土地移动距离作为相似性度量函数,以最大化视觉感知上的相似性。在目标跟踪数据集UAV20L上的实验结果表明:与现有跟踪算法相比,本算法的跟踪精确率和成功率均较高。具体而言,在car1视频中,本算法的精确率和成功率分别达到了82.9%和22.7%;在car16视频中,精确率和成功率则分别为51.4%和16.1%。说明本算法能够有效应对由车辆遮挡和快速运动所引发的目标漂移问题。在提高车辆目标跟踪的准确性方面,所提跟踪框架具有显著效果,对道路监控系统的实际应用具有重要价值。

    Abstract:

    In order to enhance the accuracy of vehicle target tracking in road surveillance, especially addressing the issue of reduced tracking precision and target loss due to changes in distance and occlusion, a vehicle target tracking framework based on an improved dingo optimization algorithm is proposed. Firstly, the framework integrates fast directional gradient histogram features and deep features to enhance the representation ability of the target vehicle. Secondly, by combining the Kalman filter initialization and a spiral update mechanism, the dingo optimization algorithm is improved to adapt to the problem of target loss during the search process. Finally, the improved dingo optimization algorithm is used as the search strategy to quickly locate the target, and the earth mover′s distance is selected as the similarity measurement function to maximize visual similarity. The experimental results on the UAV20L target tracking dataset show that, compared with existing tracking algorithms, the precision rate and success rate of this algorithm are both higher. Specifically, in the car1 video, the precision rate and success rate of this algorithm reached 82.9%and 22.7%, respectively; in the car16 video, the precision rate and success rate were 51.4% and 16.1%, respectively. This indicates that the algorithm can effectively deal with target drift caused by vehicle occlusion and rapid movement.The proposed tracking framework has a significant effect in improving the accuracy of vehicle target tracking and has important value for the practical application of road surveillance systems.

    参考文献
    相似文献
    引证文献
引用本文

王志伟,黄鹤,李文龙,杨澜,王会峰.改进野狗算法优化的车辆跟踪框架[J].哈尔滨工业大学学报,2026,58(9):27. DOI:10.11918/202411094

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2024-11-30
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-10-11
  • 出版日期:
文章二维码