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. 57, Issue 2, 21507(2020)

Real-Time Pedestrian Detection for Far-Infrared Vehicle Images and Adaptive Instance Segmentation

Yu Bo, Ma Shuhao, Li Hongyan, Li Chungeng*, and An Jubai
Author Affiliations
  • School of Information Science and Technology, Dalian Maritime University, Dalian, Liaoning 116026, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (17)
    Equations (0)
    References (27)
    Cited By (5)
    Tools
    Paper Information
    Topics
    Figures & Tables(17)
    Structural diagram of YOLO v3

    Fig. 1. Structural diagram of YOLO v3

    Download full size
    Visual representation of default anchor box

    Fig. 2. Visual representation of default anchor box

    Download full size
    Visual representation of our anchor box

    Fig. 3. Visual representation of our anchor box

    Download full size
    Target area location map

    Fig. 4. Target area location map

    Download full size
    Gray histogram of pedestrian object image

    Fig. 5. Gray histogram of pedestrian object image

    Download full size
    Denoising image

    Fig. 6. Denoising image

    Download full size
    Threshold segmentation image

    Fig. 7. Threshold segmentation image

    Download full size
    Closed operation image

    Fig. 8. Closed operation image

    Download full size
    Connected area mask

    Fig. 9. Connected area mask

    Download full size
    Instance segmentation effect diagrams for FLIR dataset. (a) Instance segmentation results; (b) manual labeling results

    Fig. 10. Instance segmentation effect diagrams for FLIR dataset. (a) Instance segmentation results; (b) manual labeling results

    Download full size
    Instance segmentation effect diagrams for our dataset. (a) Instance segmentation results; (b) manual labeling results

    Fig. 11. Instance segmentation effect diagrams for our dataset. (a) Instance segmentation results; (b) manual labeling results

    Download full size
    Comparison of results of algorithms. (a) FLIR dataset; (b) our dataset

    Fig. 12. Comparison of results of algorithms. (a) FLIR dataset; (b) our dataset

    Download full size
    P-R curves of FLIR dataset algorithm

    Fig. 13. P-R curves of FLIR dataset algorithm

    Download full size
    P-R curves of our dataset algorithm

    Fig. 14. P-R curves of our dataset algorithm

    Download full size

    • Table 1. Nine anchor size ratios

      View table

      Table 1. Nine anchor size ratios

      No.①②③④⑤⑥⑦⑧⑨
      Defaults(10,13)(16,30)(33,23)(30,61)(62,45)(59,119)(116,90)(156,198)(373,326)
      r0.770.531.430.491.380.501.290.791.14
      FLIRdatasets(3,11)(5,20)(7,14)(8,23)(10,34)(11,49)(16,52)(23,78)(41,140)
      r0.270.260.530.360.300.220.320.300.30
      Ourdatasets(9,33)(13,45)(15,61)(19,86)(21,69)(26,101)(32,133)(44,178)(70,271)
      r0.280.290.260.220.300.260.240.250.26
      Finalchooses(5,20)(10,34)(13,45)(16,52)(21,69)(23,78)(32,133)(41,140)(70,271)
      r0.260.300.290.320.300.300.240.300.26

    • Table 2. Algorithm performance evaluation table

      View table

      Table 2. Algorithm performance evaluation table

      AlgorithmBackboneFLIR datasetOur datasetInstancesegmentation
      APVelocity/(Frame·s-1)APVelocity/(Frame·s-1)
      HOG+SVM(64×128)63.1<165.2<1×
      Faster R-CNN(512×512)VGG1668.56.572.96×
      SSD(300×300)VGG1671.61673.315×
      YOLO v3(416×416)Darknet5374.43076.130×
      Our(416×416)Darknet5375.32977.628√

    • Table 3. IOU evaluation table

      View table

      Table 3. IOU evaluation table

      FLIR datasetThe first lineThe second lineThe third lineTest result for 50 images
      82.26%86.05%80.7%75%--90%
      Our datasetThe first lineThe second lineThe third lineTest result for 30 images
      86.56%85.68%82.89%70%--90%
    Tools

    Get Citation

    Copy Citation Text

    Yu Bo, Ma Shuhao, Li Hongyan, Li Chungeng, An Jubai. Real-Time Pedestrian Detection for Far-Infrared Vehicle Images and Adaptive Instance Segmentation[J]. Laser & Optoelectronics Progress, 2020, 57(2): 21507

    Download Citation

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

    Category: Machine Vision

    Received: Apr. 17, 2019

    Accepted: --

    Published Online: Jan. 3, 2020

    The Author Email: Li Chungeng (li_chungeng@dlmu.edu.cn)

    DOI:10.3788/LOP57.021507

    Topics

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