• Laser & Optoelectronics Progress
  • Vol. 57, Issue 10, 101507 (2020)
Hanbing Li*, Chunyang Xu, and Chaochao Hu
Author Affiliations
  • School of Automotive and Traffic Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China
  • show less
    DOI: 10.3788/LOP57.101507 Cite this Article Set citation alerts
    Hanbing Li, Chunyang Xu, Chaochao Hu. Improved Real-Time Vehicle Detection Method Based on YOLOV3[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101507 Copy Citation Text show less
    Inverted residual network. (a) Stride is 1; (b) stride is 2
    Fig. 1. Inverted residual network. (a) Stride is 1; (b) stride is 2
    Feature maps of different sizes in the last three layers of network. (a) 52×52; (b) 26×26; (c) 13×13
    Fig. 2. Feature maps of different sizes in the last three layers of network. (a) 52×52; (b) 26×26; (c) 13×13
    Improved network structure
    Fig. 3. Improved network structure
    P-R curves for different models
    Fig. 4. P-R curves for different models
    Model detection results in different scenarios. (a) Original images; (b) detection results of YOLOV3; (c) detection results of improved model
    Fig. 5. Model detection results in different scenarios. (a) Original images; (b) detection results of YOLOV3; (c) detection results of improved model
    ModelLayerParameterSize /MB
    SSD8827188676103
    YOLOV27550983561194
    YOLOV325661587112235
    Proposed2232231412085.6
    Table 1. Comparison of network layers and sizes of different models
    Improvement strategyInverted residualsGNSoftNMSFocal-loss
    Change of mAP /%-3.171.151.391.78
    Table 2. Influence of different improvement strategies on mAP
    ModelmAP /%Time /ms
    SSD89.8848.8
    YOLOV289.6030.2
    YOLOV391.9142.3
    Proposed93.0628.5
    Table 3. Comparison of test results of different models
    Hanbing Li, Chunyang Xu, Chaochao Hu. Improved Real-Time Vehicle Detection Method Based on YOLOV3[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101507
    Download Citation