• Laser & Optoelectronics Progress
  • Vol. 59, Issue 24, 2415002 (2022)
Lü Wei, Zhiyin Liang, and Jinghui Chu*
Author Affiliations
  • School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP202259.2415002 Cite this Article Set citation alerts
    Lü Wei, Zhiyin Liang, Jinghui Chu. Traffic Sign Detection Algorithm Based on Modified Anchor-Free Model[J]. Laser & Optoelectronics Progress, 2022, 59(24): 2415002 Copy Citation Text show less
    Proposed structure based on Anchor-Free model
    Fig. 1. Proposed structure based on Anchor-Free model
    Residual augmentation decoder module
    Fig. 2. Residual augmentation decoder module
    Structure of Ghost module
    Fig. 3. Structure of Ghost module
    Structure of Ghost bottleneck
    Fig. 4. Structure of Ghost bottleneck
    Frequency distribution of traffic signs with frequency greater than 100 in TT100K
    Fig. 5. Frequency distribution of traffic signs with frequency greater than 100 in TT100K
    P-R curves for traffic signs of different sizes
    Fig. 6. P-R curves for traffic signs of different sizes
    Visualization results
    Fig. 7. Visualization results
    MethodAll /%Small /%Medium /%Large /%Params /106Model size /MbitFLOPs /109
    Faster R-CNN5R867791921345181267
    A82738886
    YOLOv323R9287959162237802
    A85808982
    YOLOv424R9287959264214728
    A88819286
    Zhu et al22R9187948881418
    A88829191
    CenterNet-Res10111R9185959453214768
    A89789492
    CenterNet-Hourglass11R948996941917654672
    A90839593
    Proposed methodR938896951664320
    A90839593
    Table 1. Comparison results of detection performance for different sizes of traffic signs on TT100K dataset
    ModelR /%A /%AP /%Params /106Model Size /MbitFLOPs /109
    Base89.387.289.815.462.2295.4
    Base+RADM90.888.691.915.763.0296.9
    Base+MFFN92.189.492.715.863.6318.4
    Base+RADM+MFFN92.590.393.616.164.4320.1
    Table 2. Results of ablation experiments of various components on TT100K dataset