Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 9, 1296(2025)

Point cloud object detection method based on multi-pillar feature fusion

Zhihan FU1,2, Zhiyi LI3, Chuang DAI2, and Lijuan ZHANG2、*
Author Affiliations
  • 1School of Computer Science, Nanjing University of Information Science and Technology, Nanjing 210044, China
  • 2School of Internet of Things Engineering, Wuxi University, Wuxi 214105, China
  • 3College of Instrument Science and Electrical Engineering, Jilin University, Changchun 130012, China
  • show less
    Figures & Tables(9)
    Network architecture proposed in this paper
    Structure of backbone network basic block
    Visualization of detection results
    • Table 1. Comparison of 3D detection accuracy of different methods on KITTI dataset

      View table
      View in Article

      Table 1. Comparison of 3D detection accuracy of different methods on KITTI dataset

      算法模态速度/FPS总体mAP/%汽车AP/%行人AP/%骑行者AP/%
      简单中等困难简单中等困难简单中等困难
      MVX-Net24点云&彩色图像27.1063.5789.1378.8874.3755.9751.3847.2971.0653.7550.28
      F-PointNet25点云&彩色图像5.9062.3282.8968.9462.1667.4757.5050.5269.8152.6748.90
      3DSSD26点云19.40-88.8278.5877.47------
      SECOND6点云50.9265.4888.1678.7475.6651.6146.5341.5081.9864.6960.42
      MVF27点云48.6066.0287.5776.7774.2358.7653.4848.0275.1562.0558.13
      PillarNet28点云31.7065.5189.6280.5577.7650.2145.0542.2783.3662.3958.35
      SeSame-voxel29点云-67.8990.5881.4578.3353.9048.4744.1484.0166.9963.10
      CenterPoint30点云25.6366.2987.6879.3276.6154.4450.7646.3180.2462.4158.82
      VoxelNeXt31点云40.9567.1686.4577.5575.0059.6553.8748.7780.7263.1059.36
      PointPillars8点云82.7863.1187.5376.3573.2050.3943.8039.4180.0560.7156.55
      Ours (0.16 m)点云61.2065.7888.6478.8675.8454.6448.0243.2578.8063.5160.44
      Ours (0.12 m)点云55.6568.9388.0978.7975.9759.7352.8247.5482.7669.4665.23
    • Table 2. Comparison of bird’s eye view (BEV) accuracy of different methods on KITTI dataset

      View table
      View in Article

      Table 2. Comparison of bird’s eye view (BEV) accuracy of different methods on KITTI dataset

      算法模态速度/FPS总体mAP/%汽车AP/%行人AP/%骑行者AP/%
      简单中等困难简单中等困难简单中等困难
      Ours (0.12 m)点云55.6575.0291.6287.8986.8965.0558.0253.3086.5075.1270.77
      MVX-Net24点云&彩色图像27.1070.7292.8388.1785.6865.2960.2255.8875.9957.8454.59
      F-PointNet25点云&彩色图像5.9068.5287.5181.9773.5570.8261.4956.4874.5456.9053.38
      MVF27点云48.6071.6690.2686.7785.6163.0058.2551.8979.2666.9163.01
      SECOND6点云50.9271.5391.9987.8686.5355.7851.3646.9385.9570.8666.55
      PillarNet28点云31.7071.5693.8787.7486.6357.5652.4349.3986.2867.1863.00
      CenterPoint30点云25.6371.8591.3288.2686.2460.5156.8353.0283.1565.4361.84
      VoxelNeXt31点云40.9573.7491.7387.4585.0165.2759.6954.6786.8368.2764.76
      PointRCNN4点云12.2274.8789.1488.6086.4965.5158.3451.6887.8973.9072.28
      PointPillars8点云82.7869.1492.4287.8685.1955.9849.6245.1182.7763.7759.53
    • Table 3. Comparison of 3D detection accuracy of different methods on DAIR-V2X dataset

      View table
      View in Article

      Table 3. Comparison of 3D detection accuracy of different methods on DAIR-V2X dataset

      算法总体mAP/%mAP(中等)/%汽车AP/%行人AP/%骑行者AP/%
      简单中等困难简单中等困难简单中等困难
      Ours48.7347.2569.2357.4954.6344.3441.4940.4647.1142.7641.02
      MVF2742.5640.8362.9650.4947.5641.1537.6236.6139.8934.3732.42
      SECOND645.7244.4668.6656.8753.8938.7436.8136.1242.6239.7038.04
      PillarNet2844.5843.2068.9057.2054.0137.8935.5735.1041.0436.8234.66
      CenterPoint3050.9149.8171.1358.9355.7548.6045.3644.6344.4645.1544.21
      VoxelNeXt3152.6451.4170.0058.0754.8852.8448.7647.5147.7847.4146.49
      PointRCNN443.6541.5471.3757.5454.3038.9332.7731.5139.6434.3032.52
      PointPillars839.3237.9567.0055.0951.8631.9730.5130.1931.9928.2527.03
    • Table 4. Comparison of bird’s eye view (BEV) accuracy of different methods on DAIR-V2X dataset

      View table
      View in Article

      Table 4. Comparison of bird’s eye view (BEV) accuracy of different methods on DAIR-V2X dataset

      算法总体mAP/%mAP(中等)/%汽车AP/%行人AP/%骑行者AP/%
      简单中等困难简单中等困难简单中等困难
      Ours50.9249.5771.8560.2057.6146.0043.3142.4248.0945.2043.60
      MVF2744.6843.0365.4953.1750.4642.9539.7439.1540.7836.1934.14
      SECOND648.0246.8871.4359.7757.2139.9038.6838.2543.9142.1940.86
      PillarNet2847.0145.7771.7860.1057.4339.4137.9137.7442.0539.3037.35
      CenterPoint3051.8550.7673.4661.4158.8245.1445.4345.0047.1545.4444.78
      VoxelNeXt3155.8754.8173.0361.0758.3655.2252.0251.0350.4251.3550.34
      PointRCNN444.0142.3272.3360.0655.1138.4833.3132.2138.9533.5932.04
      PointPillars841.9840.9771.2159.8157.1534.1633.0030.1933.3730.1028.87
    • Table 5. Ablation study on KITTI dataset

      View table
      View in Article

      Table 5. Ablation study on KITTI dataset

      算法中等难度AP/%mAP/%速度/FPS
      柱体编码器BBDACarPed.Cyc.
      MFXB
      PointPillars 876.343.559.159.687.4
      77.842.160.760.270.8
      76.444.762.161.185.8
      76.243.863.461.273.7
      76.346.562.161.688.3
      78.442.863.761.661.8
      76.747.561.661.968.7
      76.746.266.463.173.0
      78.948.063.563.565.5
    • Table 6. Ablation study of data augmentation on KITTI dataset

      View table
      View in Article

      Table 6. Ablation study of data augmentation on KITTI dataset

      算法中等难度AP/%mAP/%
      CarPed.Cyc.
      PA-AUG1676.4447.0362.2261.90
      Ours77.8148.4162.4762.90
    Tools

    Get Citation

    Copy Citation Text

    Zhihan FU, Zhiyi LI, Chuang DAI, Lijuan ZHANG. Point cloud object detection method based on multi-pillar feature fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(9): 1296

    Download Citation

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

    Category:

    Received: Apr. 25, 2025

    Accepted: --

    Published Online: Sep. 25, 2025

    The Author Email: Lijuan ZHANG (zhanglijuan@cwxu.edu.cn)

    DOI:10.37188/CJLCD.2025-0096

    Topics