Acta Optica Sinica, Volume. 45, Issue 9, 0910002(2025)

Infrared Traffic Object Detection Network for Edge Device Deployment

Yulan Han*, Deao Chen, Tong Wu, Xianlu Liu, and Chaofeng Lan
Author Affiliations
  • Heilongjiang Province Key Laboratory of Pattern Recognition and Information Perception, School of Measurement and Control Technology and Communication Engineering, Harbin University of Science and Technology, Harbin 150080, Heilongjiang , China
  • show less
    Figures & Tables(10)
    Edge-DETR network model. (a) Overall framework diagram; (b) GISM; (c) zoom cat block
    CAA-MELAN block
    HiLo block
    CFFM structure
    Visualization images of infrared feature extraction by backbone network. (a) Features extracted by RT-DETR; (b) features extracted by Edge-DETR
    Heatmaps of detection results. (a) Heatmaps of detection results by RT-DETR; (b) heatmaps of detection results by Edge-DETR
    Comparison of detection effects on three datasets by different algorithms. (a) RGB images; (b) RGB detection images; (c) infrared images; (d) Fast R-CNN; (e) YOLOv11; (f) RT-DETR; (g) ours
    Comparison of detection effects on self-built dataset by different algorithms. (a) RGB images; (b) RGB detection images; (c) infrared images; (d) Fast R-CNN; (e) YOLOv11; (f) RT-DETR; (g) ours
    • Table 1. Ablation experiments on FLIR

      View table

      Table 1. Ablation experiments on FLIR

      GroupCAA-MELANCFFMHiLoShape-IoUmAP50 /%mAP75 /%Params /106FLOPs /109Size /MBFR /(frame/s)
      149.3929.6019.8756.938.651.3
      251.9230.8612.1438.623.243.6
      350.5630.3219.8659.839.541.4
      450.2130.7319.8656.826.557.2
      550.3630.3219.8756.938.651.3
      650.2330.7419.8748.930.641.9
      752.6731.4710.6934.721.036.7
      853.8932.3410.6934.721.036.7
    • Table 2. Comparative experiment results on different datasets

      View table

      Table 2. Comparative experiment results on different datasets

      ModelLLVIPFLIRKAISTSelf-built datasetFLOPs /109Params /106Size /MB
      mAP50 /%mAP75 /%mAP50 /%mAP75 /%mAP50 /%mAP75 /%mAP50 /%mAP75 /%
      SSD591.1256.2720.037.5441.4918.1481.6862.5162.726.20100
      Fast R-CNN696.1468.1623.428.4141.5111.8583.8167.60177.541.30129
      YOLOv5m894.6172.2144.2027.5047.5634.5182.2265.94114.546.6081
      YOLOv8m996.2276.5548.1030.7147.1335.5583.5466.8878.725.8092
      YOLOv10m1096.5578.2350.3731.0947.8335.8683.8463.3959.115.8063
      YOLOv11m1196.4978.9251.3231.2647.7236.1483.9664.7967.920.1038
      DETR-R501395.4376.3346.7629.2444.3136.4382.1163.3086.641.70171

      Deformable

      DETR14

      96.4878.5449.3330.4746.9634.4183.0365.3742.227.10183
      RT-DETR1596.5779.8549.3929.6048.1436.2783.1966.3156.919.8738
      Edge-DETR(ours)97.9882.5553.8932.3448.2836.5584.3068.4234.710.6921
    Tools

    Get Citation

    Copy Citation Text

    Yulan Han, Deao Chen, Tong Wu, Xianlu Liu, Chaofeng Lan. Infrared Traffic Object Detection Network for Edge Device Deployment[J]. Acta Optica Sinica, 2025, 45(9): 0910002

    Download Citation

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

    Category: Image Processing

    Received: Dec. 19, 2024

    Accepted: Mar. 11, 2025

    Published Online: May. 20, 2025

    The Author Email: Yulan Han (hanyulan@hrbust.edu.cn)

    DOI:10.3788/AOS241913

    CSTR:32393.14.AOS241913

    Topics