Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 58, Issue 6, 615002(2021)

Research on Helmet Wear Identification Based on Improved YOLOv3

Xu Kai and Deng Chao*
Author Affiliations
  • School of Physics & Electronic Information Engineering, Henan Polytechnic University, Jiaozuo, Henan454003,China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (11)
    Equations (0)
    References (21)
    Cited By (3)
    Tools
    Paper Information
    Topics
    Figures & Tables(11)
    YOLOv3 model structure

    Fig. 1. YOLOv3 model structure

    Download full size
    Safety helmet wear identification process

    Fig. 2. Safety helmet wear identification process

    Download full size
    Diagram of YOLOv3-4l model

    Fig. 3. Diagram of YOLOv3-4l model

    Download full size
    K-means cluster analysis results

    Fig. 4. K-means cluster analysis results

    Download full size
    Comparison chart of IoU standard and GIoU standard. (a) L2 norm; (b) L1 norm

    Fig. 5. Comparison chart of IoU standard and GIoU standard. (a) L2 norm; (b) L1 norm

    Download full size
    Change of loss value

    Fig. 6. Change of loss value

    Download full size
    Test results. (a)(b) Intensive targets; (c)(d) small targets

    Fig. 7. Test results. (a)(b) Intensive targets; (c)(d) small targets

    Download full size

    • Table 1. Distribution of a priori candidate box after clustering

      View table

      Table 1. Distribution of a priori candidate box after clustering

      Feature map sizeCandidate box size
      104×1048, 1711, 2316, 31
      52×5224, 4134, 5646, 80
      26×2666, 10582, 152150, 157
      13×13112, 225185, 304335, 427

    • Table 2. Data set distribution

      View table

      Table 2. Data set distribution

      ClassMarked numberTrain set numberTest set number
      Hat69176343571
      Person85962781047858

    • Table 3. Comparison of detection accuracy of different models

      View table

      Table 3. Comparison of detection accuracy of different models

      GroupAlgorithmHatPersonmAPTime /sSpeed /(frame·s-1)
      1YOLOv30.87570.85680.86620.02245
      2YOLOv3-4l0.89720.84390.87050.02540
      3YOLOv3-4l+GIoU Loss0.92060.81220.86640.02934
      4YOLOv3-4l+Focal Loss0.89300.81830.85560.03033
      5GF-YOLOv3(proposed)0.91800.88370.90090.02934

    • Table 4. Wear identification results statistics

      View table

      Table 4. Wear identification results statistics

      GroupAlgorithmTotalPTPPFPPFNPrecisionRecall
      1YOLOv35006742100710970.870.86
      2YOLOv3-4l500663510804990.860.93
      3YOLOv3-4l+GIoU Loss500671899412360.870.84
      4YOLOv3-4l+Focal Loss500664144713130.940.83
      5GF-YOLOv3(proposed)50067284293540.940.95
    Tools

    Get Citation

    Copy Citation Text

    Xu Kai, Deng Chao. Research on Helmet Wear Identification Based on Improved YOLOv3[J]. Laser & Optoelectronics Progress, 2021, 58(6): 615002

    Download Citation

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

    Category: Machine Vision

    Received: Jul. 24, 2020

    Accepted: --

    Published Online: Mar. 6, 2021

    The Author Email: Deng Chao (super@hpu.edu.cn)

    DOI:10.3788/LOP202158.0615002

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology