基于中心点的多类别车辆检测算法
DOI:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

TP391

基金项目:

国家自然科学基金(51365017,61463018);江西省自然科学基金面上项目(20192BAB205084);江西省教育厅科学技术研究重点项目(GJJ170491)。


Center Based algorithm for multi-class vehicle detection
Author:
Affiliation:

Fund Project:

The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)

  • 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:

    针对多类别车辆检测任务中存在计算复杂,检测精度不高的问题,提出一种基于中心点的多类别车辆检测算法。该算法首先通过Hourglass网络对各类型车辆特征进行提取,考虑到多类别车辆检测时易受车辆大小、视觉变化及非刚体形变等因素的影响,采用可变形卷积替换传统卷积的方法对Hourglass网络重建;在网络预测模块中,结合不同的预测分支支路,采用组合损失函数度量模型拟合的程度,同时引入GIoU损失提高模型拟合效果,减少车辆检测中漏检和误检现象的发生;最后通过Sigmoid激活函数得到最终的检测结果。在公开数据库上仿真实验,测试精度和检测速度分别达到了93.42%和49f/s,在自制数据库上仿真实验,所提算法的精确率和召回率相比CenterNet算法分别提高了2.7% 和5.6%。实验结果表明,本文算法在车辆检测任务中具有明显优势。

    Abstract:

    In order to solve the problem of complex calculation and low detection accuracy in multi-class vehicle detection, a multi-class vehicle detection algorithm based on center point was proposed. Firstly, the algorithm extracted the characteristics of multi-class vehicles through the Hourglass network. Considering that the detection of multi-class vehicles was susceptible to the influence of vehicle size, visual change and non-rigid body deformation, the method of deformable convolution was used to replace the traditional convolution to reconstruct the Hourglass network. The combined loss function was used to measure the degree of model fitting combined with different branches of prediction in the network prediction module, and GIoU loss was introduced to improve the model fitting effect, so as to reduce the occurrence of missed and false checks in vehicle detection. Finally, the Sigmoid activation function was used to obtain the final detection result. The simulation experiments shows that the testing accuracy and speed reache 93.42% and 49 f/s on public database. The precision rate and recall rate of the proposed algorithm is improved by 2.7% and 5.6% compared with CenterNet on homemade database. The results shows that this algorithm can perform better in the vehicle detection task.

    参考文献
    相似文献
    引证文献
引用本文

梁礼明,熊文,彭仁杰,等. 基于中心点的多类别车辆检测算法[J]. 科学技术与工程, 2021, 21(7): 2767-2772.
Liang Liming, Xiong Wen, Peng Renjie, et al. Center Based algorithm for multi-class vehicle detection[J]. Science Technology and Engineering,2021,21(7):2767-2772.

复制
分享
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:2020-05-27
  • 最后修改日期:2020-12-17
  • 录用日期:2020-09-16
  • 在线发布日期: 2021-03-31
  • 出版日期:
×
律回春渐,新元肇启|《科学技术与工程》编辑部恭祝新岁!
亟待确认版面费归属稿件,敬请作者关注