Optics and Precision Engineering, Volume. 31, Issue 9, 1366(2023)

Road traffic sign recognition algorithm based on improved YOLOv4

Daxiang LI, Zhongheng SU*, and Ying LIU
Author Affiliations
  • College of Communication and Information Engineering, Xi'an University of Posts and Telecommunication, Xi'an710121, China
  • show less
    Figures & Tables(12)
    Network structure of YOLOv4
    Flow chart of YOLOv4 object recognition
    Network structure of improved YOLOv4
    Attention-driven scale-aware module
    Feature-aligned pyramid convolution module
    Feature-aligned module
    Size distribution of traffic signs from TT100K
    Recognition results of improved YOLOv4 and original YOLOv4
    • Table 1. Experimental software and hardware configuration

      View table
      View in Article

      Table 1. Experimental software and hardware configuration

      类别名称型号和参数

      硬件

      中央处理器Intel(R)Core(TM) Xeon E5-2640
      内存128GB
      图像处理器NVIDIA Titan X(12G)

      软件

      操作系统Ubuntu 18.04
      深度学习框架Pytorch-cuda 1.7.0
      开发语言Python 3.8.0
      环境管理Anaconda 3.6.5
    • Table 2. Anchor size generated by K-means++ algorithm

      View table
      View in Article

      Table 2. Anchor size generated by K-means++ algorithm

      Feature mapReceptive fieldAnchor box
      19×19Large

      (74,80)

      (94,100)

      (140,146)

      38×38Medium

      (37,40)

      (46,52)

      (58,63)

      76×76Small

      (14,18)

      (22,24)

      (28,32)

    • Table 3. Performance comparison of different methods on TT100K dataset

      View table
      View in Article

      Table 3. Performance comparison of different methods on TT100K dataset

      MethodsSmallMediumLargeOverall
      PRFPRFPRFPRF
      Zhu et al.3281.787.484.590.893.692.290.687.789.187.789.688.6
      Faster R-CNN3325.258.235.263.883.772.480.791.285.656.777.665.5
      YOLOv42083.087.585.291.395.293.290.688.289.488.390.389.3
      FAMN3488.490.189.294.297.295.792.896.194.491.894.393.0
      DR-CNN3583.189.386.191.794.893.292.489.691.089.091.290.1
      Noh et al.[36]84.892.688.594.297.595.893.397.595.490.695.793.1
      Wang et al.[37]87.389.488.392.596.494.492.890.591.690.892.391.5
      TsingNet1589.090.689.895.295.695.496.292.894.593.492.993.1
      Ours89.492.390.895.596.896.196.497.496.993.794.594.1
    • Table 4. Ablation experiment data

      View table
      View in Article

      Table 4. Ablation experiment data

      MethodsADSAFAPCPsPmPlPallFPS
      YOLOv4 baseline--83.091.390.688.341.36
      -87.392.793.491.237.74
      -86.693.894.591.638.32
      Ours89.495.596.493.733.17
    Tools

    Get Citation

    Copy Citation Text

    Daxiang LI, Zhongheng SU, Ying LIU. Road traffic sign recognition algorithm based on improved YOLOv4[J]. Optics and Precision Engineering, 2023, 31(9): 1366

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Information Sciences

    Received: Jul. 13, 2022

    Accepted: --

    Published Online: Jun. 6, 2023

    The Author Email: Zhongheng SU (Szh1998@stu.xupt.edu.cn)

    DOI:10.37188/OPE.20233109.1366

    Topics