Optics and Precision Engineering, Volume. 32, Issue 10, 1595(2024)

A method for dense occlusion target recognition of service robots based on improved YOLOv7

Renxiang CHEN1,*... Tianran QIU1, Lixia YANG2, Tenwei YU1, Fei JIA1 and Cai CHEN3 |Show fewer author(s)
Author Affiliations
  • 1Chongqing Engineering Laboratory of Traffic Engineering Application Robot, Chongqing Jiaotong University,Chongqing400074, China
  • 2School of Business Administration, Chongqing University of Science and Technology, Chongqing401331, China
  • 3Chongqing Intelligent Robot Research Institute, Chongqing 4000714, China
  • show less
    Figures & Tables(14)
    Improve YOLOv7 structure
    Depthwise Over-parameterized module
    Coordinate attention module
    Overall flowchart of Recognition
    Model recognition visualization results
    No dense occlusion to recognize visualization results
    Service robot experimental platform
    Comparison of actual recognition effects
    Dense occlusion images from MessyTable dataset
    • Table 1. Dense occlusion of dataset

      View table
      View in Article

      Table 1. Dense occlusion of dataset

      无密集遮挡样本个数密集遮挡样本个数
      训练集1 1882 264
      验证集108230
      测试集100166
    • Table 2. Results of ablation experiment

      View table
      View in Article

      Table 2. Results of ablation experiment

      模型P/%R/%mAP/%模型体积/MBFPS/(frame·s-1
      YOLOv787.982.888.87231.3
      改进186.582.891.17429.5
      改进286.588.391.47230.2
      改进382.185.389.24838.0
      改进1+293.086.092.57428.6
      改进1+393.986.491.65036.4
      改进2+390.084.091.54837.0
      Im-YOLOv792.888.792.95035.8
    • Table 3. No dense occlusion test data

      View table
      View in Article

      Table 3. No dense occlusion test data

      测试数据方法P/%R/%mAP/%
      无遮挡场景YOLOv795.792.495.6
      无遮挡场景Im-YOLOv792.295.097.5
    • Table 4. Self built dataset dataset comparison experiment results

      View table
      View in Article

      Table 4. Self built dataset dataset comparison experiment results

      模型P/%R/%mAP/%模型体积/MBFPS/(frame·s-1
      DETR580.175.485.64738.9
      YOLOv5-s85.378.086.31432.2
      YOLOv5改进2190.276.086.73628.6
      YOLOv5-l90.481.288.59030.2
      YOLOv4-tiny-x889.182.988.68234.4
      YOLOv787.982.888.87231.3
      YOLOv8-l92.378.489.68434.1
      Im-YOLOv792.888.792.95035.8
    • Table 5. MessyTable dataset comparison experiment results

      View table
      View in Article

      Table 5. MessyTable dataset comparison experiment results

      模型P/%R/%mAP/%模型体积/MB
      DETR583.474.980.6473
      YOLOv5-s85.772.481.214
      YOLOv5改进2186.276.682.236
      YOLOv5-l90.074.883.090
      YOLOv4-tiny-x885.574.484.482
      YOLOv782.881.184.672
      YOLOv8-l89.671.984.184
      Im-YOLOv784.783.387.850
    Tools

    Get Citation

    Copy Citation Text

    Renxiang CHEN, Tianran QIU, Lixia YANG, Tenwei YU, Fei JIA, Cai CHEN. A method for dense occlusion target recognition of service robots based on improved YOLOv7[J]. Optics and Precision Engineering, 2024, 32(10): 1595

    Download Citation

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

    Category:

    Received: Nov. 14, 2023

    Accepted: --

    Published Online: Jul. 8, 2024

    The Author Email: CHEN Renxiang (manlou.yue@126.com)

    DOI:10.37188/OPE.20243210.1595

    Topics