Acta Photonica Sinica, Volume. 53, Issue 1, 0111002(2024)

BEV Space 3D Object Detection Algorithm Based on Fusion of Infrared Camera and LiDAR

Wuyue WANG1、*, Zhaofei XU2,3, Chunyan QU3, Ying LIN4, Yufeng CHEN4, and Jian LIAO1
Author Affiliations
  • 1Yantai Research Institute,Harbin Engineering University,Yantai 265500,China
  • 2Mechanical and Electrical Engineering Institute,Harbin Engineering University,Harbin 150000,China
  • 3Iray Optoelectronic Technology Co.,LTD,Yantai 265500,China
  • 4Electric Power Research Institute,State Grid Shandong Electric Power Company,Jinan 250014,China
  • show less
    Figures & Tables(16)
    Frame figure of separable fusion sensing system
    Image encoder module
    DepthNet
    Improved DepthNet
    BEV pooling accelerator kernel
    Lidar branch structure
    Gating attention fusion mechanism module
    Data collection hardware platform and layout position
    Dataset scene distribution
    Embedded AI computing platform MIIVII APEX AD10
    City road scene test(RVIZ)
    • Table 1. Ablation experiment of camera branch

      View table
      View in Article

      Table 1. Ablation experiment of camera branch

      DSVCPPBEV mAP/%3D mAP/%AOS/%
      26.6923.4446.74
      30.2725.1657.92
      31.8925.9559.31
    • Table 2. Ablation experiment of fusion branch

      View table
      View in Article

      Table 2. Ablation experiment of fusion branch

      GAFBEV mAP/%3D mAP/%AOS/%
      76.7266.9569.30
      77.1468.0870.19
    • Table 3. Performance comparison of different voxelization method in lidar branches

      View table
      View in Article

      Table 3. Performance comparison of different voxelization method in lidar branches

      V methodBEV AP/%3D AP/%AOS/%
      Car pedestrian cyclistCar pedestrian cyclistCar pedestrian cyclist
      SV89.9177.2980.3187.7677.0980.2385.5757.1476.97
      DV90.2078.1980.1588.6778.1380.0985.4858.8377.41
    • Table 4. Performance comparison of camera branch,lidar branch and fusion branch

      View table
      View in Article

      Table 4. Performance comparison of camera branch,lidar branch and fusion branch

      ModalityDSVCPPDVGAFBEV AP/%3D AP/%AOS/%
      Car pedestrian cyclistCar pedestrian cyclistCar pedestrian cyclist
      Camera64.6731.5557.1958.4629.7155.6666.4544.6966.80
      Lidar90.2078.1980.1588.6778.1380.0985.4858.8377.41
      Lidar & camera91.2577.3284.0690.8477.1582.8885.8060.1980.46
    • Table 5. Performance comparison of our model with other SOTA model

      View table
      View in Article

      Table 5. Performance comparison of our model with other SOTA model

      MethodModalityBEV AP/%3D AP/%AOS/%
      Car pedestrian cyclistCar pedestrian cyclistCar pedestrian cyclist
      PointPillars17Lidar81.7268.3678.4681.6568.3578.4279.6651.9375.31
      CenterPoint18Lidar90.2078.1980.1588.6778.1380.0985.4858.8377.41
      MVXNet24Lidar & camera89.1976.1673.1689.0176.1371.7182.9149.0561.03
      OursLidar & camera91.2577.3284.0690.8477.1582.8885.8060.1980.46
    Tools

    Get Citation

    Copy Citation Text

    Wuyue WANG, Zhaofei XU, Chunyan QU, Ying LIN, Yufeng CHEN, Jian LIAO. BEV Space 3D Object Detection Algorithm Based on Fusion of Infrared Camera and LiDAR[J]. Acta Photonica Sinica, 2024, 53(1): 0111002

    Download Citation

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

    Category:

    Received: Jul. 10, 2023

    Accepted: Sep. 18, 2023

    Published Online: Feb. 1, 2024

    The Author Email: Wuyue WANG (wangwuyue@hrbeu.edu.cn)

    DOI:10.3788/gzxb20245301.0111002

    Topics