Optics and Precision Engineering, Volume. 31, Issue 10, 1532(2023)

Dense pedestrian detection algorithm in multi-branch non-anchor frame network

Zhixuan LÜ... Xia WEI* and Deqi HUANG |Show fewer author(s)
Author Affiliations
  • School of Electrical Engineering, XinJiang University, Urumqi830049, China
  • show less
    Figures & Tables(17)
    MBAN model architecture
    Feature Extraction Net
    Local feature point change extracted by FEN
    Results of prediction framework output
    Learning rate changes during training
    Loss changes of the model during training
    Loss changes of the global during training
    Loss changes of the head and legs during training
    Network structure of baseline model
    Prediction results of CenterNet model
    Prediction results of MBAN model
    Prediction results of MBAN model with FSN
    • Table 1. Experimental training environment configuration

      View table
      View in Article

      Table 1. Experimental training environment configuration

      名称配置
      CPUIntel Xeon E5-2678 v3
      显卡NVIDIA Tesla K80
      加速器CUDA,CUDNN
    • Table 2. Result of system scale calibration

      View table
      View in Article

      Table 2. Result of system scale calibration

      MethodsmAP/%F1Prec/%Recall/%FPSParams
      Baseline32.180.4741.7053.3114.3932.78
      Baseline+hourglass+1FEN(global)61.450.6768.9365.5013.3832.78
      Baseline+hourglass+2FEN(global+head)68.320.7875.3179.8612.4232.78
      Baseline+hourglass+3FEN(global+head+legs)75.040.8378.8986.5511.1932.78
      Baseline+hourglass+3FEN+LFD85.220.8780.0794.3911.1032.79
    • Table 3. Visualization of each branch output of MBAN

      View table
      View in Article

      Table 3. Visualization of each branch output of MBAN

      原图
      F1
      Fg
      F2
    • Table 4. Fig.4 Output visual of different network structure models

      View table
      View in Article

      Table 4. Fig.4 Output visual of different network structure models

      原图
      Baseline
      Baseline+2FEN
      Baseline+3FEN+LFD(MBAN)
    • Table 5. Comparison results between MBAN method and other methods

      View table
      View in Article

      Table 5. Comparison results between MBAN method and other methods

      MethodsBackboneSizemAP/%Params(×106GFLOPsFPS
      VGG16-SSDVGG16512×51269.1660.322.935
      YOLOv3Darknet53640×64084.2261.5154.927
      YOLOv3-tinyDarknet19640×64068.238.012.9100
      ResNet50-FCOSResnet50640×64069.6650.2195.922
      CenterNetHourglass104512×51263.1625.722.114.
      MBAN(CrowHuman)Resnet50512×51286.9832.823.811
      MBAN(WiderPerson)Resnet50512×51285.2232.823.811
    Tools

    Get Citation

    Copy Citation Text

    Zhixuan LÜ, Xia WEI, Deqi HUANG. Dense pedestrian detection algorithm in multi-branch non-anchor frame network[J]. Optics and Precision Engineering, 2023, 31(10): 1532

    Download Citation

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

    Category: Information Sciences

    Received: May. 30, 2022

    Accepted: --

    Published Online: Jul. 4, 2023

    The Author Email: WEI Xia (30462111@qq.com)

    DOI:10.37188/OPE.20233110.1532

    Topics