《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (7): 2219-2226.DOI: 10.11772/j.issn.1001-9081.2021050731

• 多媒体计算与计算机仿真 • 上一篇    

基于改进YOLOv3的实时交通标志检测算法

张达为1(), 刘绪崇2, 周维1, 陈柱辉1, 余瑶3   

  1. 1.湘潭大学 计算机学院·网络空间安全学院, 湘潭 湖南, 411105
    2.湖南警察学院 湖南公安科学技术研究院, 长沙 410138
    3.湘潭大学 公共管理学院, 湘潭 湖南, 411105
  • 收稿日期:2021-05-10 修回日期:2021-10-31 接受日期:2021-11-08 发布日期:2022-07-15 出版日期:2022-07-10
  • 通讯作者: 张达为
  • 作者简介:刘绪崇(1973—),男,湖南桑植人,教授,博士,CCF会员,主要研究方向:大数据分析、信息网络安全
    周维(1978—),男,湖南湘潭人,副教授,博士,CCF会员,主要研究方向:计算机视觉、智能系统
    陈柱辉(1996—),男,湖南永州人,硕士研究生,CCF会员,主要研究方向:自然语言处理
    余瑶(1995—),女,江西上饶人,硕士研究生,主要研究方向:公共舆论分析。
  • 基金资助:
    湖南省自然科学基金资助项目(2018JJ2107);湖南省科技重大专项(2017SK1040);湖南省高新技术产业科技创新引领计划项目(2020GK2029)

Real-time traffic sign detection algorithm based on improved YOLOv3

Dawei ZHANG1(), Xuchong LIU2, Wei ZHOU1, Zhuhui CHEN1, Yao YU3   

  1. 1.School of Computer Science & School of Cyberspace Science,Xiangtan University,Xiangtan Hunan 411105,China
    2.Hunan Academy of Public Security Science and Technology,Hunan Police Academy,Changsha Hunan 410138,China
    3.School of Public Administration,Xiangtan University,Xiangtan Hunan 411105,China
  • Received:2021-05-10 Revised:2021-10-31 Accepted:2021-11-08 Online:2022-07-15 Published:2022-07-10
  • Contact: Dawei ZHANG
  • About author:LIU Xuchong, born in 1973, Ph. D., professor. His research interests include big data analysis, information network security.
    ZHOU Wei, born in 1978, Ph. D., associate professor. His research interests include computer vision, intelligent systems.
    CHEN Zhuhui, born in 1996, M. S. candidate. His research interests include natural language processing.
    YU Yao, born in 1995, M. S. candidate. Her research interests include public opinion analysis.
  • Supported by:
    Hunan Provincial Natural Science Foundation(2018JJ2107);Major Science and Technology Project of Hunan Province(2017SK1040);Project of Hunan Province High-tech Industry Science and Technology Innovation Leading Plan(2020GK2029)

摘要:

针对目前我国智能驾驶辅助系统识别道路交通标志检测速度慢、识别精度低等问题,提出一种基于YOLOv3的改进的道路交通标志检测算法。首先,将MobileNetv2作为基础特征提取网络引入YOLOv3以形成目标检测网络模块MN-YOLOv3,在MN-YOLOv3主干网络中引入两条Down-up连接进行特征融合,从而减少检测算法的模型参数,提高了检测模块的运行速度,增强了多尺度特征图之间的信息融合;然后,根据交通标志目标形状的特点,使用K-Means++算法产生先验框的初始聚类中心,并在边界框回归中引入距离交并比(DIOU)损失函数来将DIOU与非极大值抑制(NMS)结合;最后,将感兴趣区域(ROI)与上下文信息通过ROI Align统一尺寸后融合,从而增强目标特征表达。实验结果表明,所提算法性能更好,在长沙理工大学中国交通标志检测(CCTSDB)数据集上的平均准确率均值(mAP)可达96.20%。相较于Faster R-CNN、YOLOv3、Cascaded R-CNN检测算法,所提算法拥有具有更好的实时性和更高的检测精度,对各种环境变化具有更好的鲁棒性。

关键词: 目标检测, 特征融合, YOLOv3, 距离交并比, MobileNetv2, K-Means++

Abstract:

Aiming at the problems of slow detection and low recognition accuracy of road traffic signs in Chinese intelligent driving assistance system, an improved road traffic sign detection algorithm based on YOLOv3 (You Only Look Once version 3) was proposed. Firstly, MobileNetv2 was introduced into YOLOv3 as the basic feature extraction network to construct an object detection network module MN-YOLOv3 (MobileNetv2-YOLOv3). And two Down-up links were added to the backbone network of MN-YOLOv3 for feature fusion, thereby reducing the model parameters, and improving the running speed of the detection module as well as information fusion performance of the multi-scale feature maps. Then, according to the shape characteristics of traffic sign objects, K-Means++ algorithm was used to generate the initial cluster center of the anchor, and the DIOU (Distance Intersection Over Union) loss function was introduced to combine DIOU and Non-Maximum Suppression (NMS) for the bounding box regression. Finally, the Region Of Interest (ROI) and the context information were unified by ROI Align and merged to enhance the object feature expression. Experimental results show that the proposed algorithm has better performance, and the mean Average Precision (mAP) of the algorithm on the dataset CSUST (ChangSha University of Science and Technology) Chinese Traffic Sign Detection Benchmark (CCTSDB) can reach 96.20%. Compared with Faster R-CNN (Region Convolutional Neural Network), YOLOv3 and Cascaded R-CNN detection algorithms, the proposed algorithm has better real-time performance, higher detection accuracy, and is more robustness to various environmental changes.

Key words: object detection, feature fusion, You Only Look Once version 3 (YOLOv3), DIOU (Distance Intersection Over Union), MobileNetv2, K-Means++

中图分类号: