Chinese Journal of Lasers, Volume. 51, Issue 13, 1310003(2024)

Point‑Voxel Consistency Constraint Network for LiDAR Point Cloud Classification Under Urban Scenes

Huchen Li, Haiyan Guan*, Xiangda Lei, Nannan Qin, and Huan Ni
Author Affiliations
  • School of Remote Sensing & Geomatics Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, Jiangsu , China
  • show less
    Figures & Tables(16)
    Overall structure of the PVCC-Net
    Point-voxel consistency constraint module
    Point-voxel self-attention fusion module
    Overall visualization of the classification results on the Toronto3D dataset. (a) Point cloud (RGB); (b) ground truth; (c) PVCC-Net; (d) RandLA-Net; (e) PVCNN
    Local visualization of classification results on the Toronto3D dataset. (a) Point cloud (RGB); (b) ground truth; (c) PVCC-Net; (d) RandLA-Net; (e) PVCNN
    OA and mIoU of different methods on the Semantic3D dataset. (a) OA; (b) mIoU
    Visualization of classification results on the Semantic3D dataset. (a) Point cloud (RGB); (b) PVCC-Net; (c) RandLA-Net;
    Normalized confusion matrix of classification results on the Semantic3D dataset
    OA and mIoU of different methods on the SensatUrban dataset. (a) OA; (b) mIoU
    Visualization of classification results on the SensatUrban dataset. (a) Point cloud (RGB); (b) PVCC-Net; (c) RandLA-Net;
    • Table 1. Introduction of point cloud classification dataset in urban scenes

      View table

      Table 1. Introduction of point cloud classification dataset in urban scenes

      DatasetRegionSpatial sizeNumber of points/106RGBNumber of classesSensor
      Toronto3DToronto, Canada1×103 m78.3Yes8MLS
      Semantic3DCentral Europe3.84×104 m24000Yes8TLS
      SensatUrbanBirmingham, Cambridge and York, UK7.64×106 m22847Yes13ALS
    • Table 2. Comparison of classification results of different methods on Toronto3D dataset

      View table

      Table 2. Comparison of classification results of different methods on Toronto3D dataset

      MethodOAmIoUIoUs
      roadroad marknaturalbuildingutil. linepolecarfence
      ResDLPS-Net96.4980.2795.8259.8096.1090.9686.8279.9589.4143.31
      BAAF-Net94.2081.2096.8067.3096.8092.2086.8082.3093.1034.00
      BAF-LAC95.2082.2096.6064.7096.4092.8086.1083.9093.7043.50
      MFA97.0079.9096.8070.0096.1092.3086.3080.4091.5029.40
      LACV-Net97.4082.7097.1066.9097.3093.0087.3083.4093.4043.10
      RandLA-Net94.3781.7796.6964.2196.9294.2488.0677.8493.3742.86
      MappingConvSeg94.7282.8997.1567.8797.5593.7586.8882.1293.7244.11
      PVCNN*97.3678.5697.6871.3194.8191.9381.8777.4290.5622.90
      Baseline*96.4479.4895.9060.5995.7691.3186.0579.1389.6037.53
      PVCC-Net*97.9782.9297.8774.9597.0093.9186.6581.7194.8636.42
    • Table 3. Comparison of classification results of different methods on Semantic3D dataset

      View table

      Table 3. Comparison of classification results of different methods on Semantic3D dataset

      MethodIoUs /%
      man-made.natural.high veg.low veg.buildinghardscapescanning art.car
      SnapNet82.0077.3079.7022.9091.1018.4037.3064.40
      DeepVoxNet82.7053.1083.8028.7089.9023.6029.8065.00
      MSDeepVoxNet83.0067.2083.9036.7092.4031.3050.0078.20
      OctNet90.7082.0082.4039.3090.0010.9031.2046.00
      SEGCloud83.9066.0086.0040.5091.1030.9027.5064.30
      RFMSSF87.6080.3081.8036.4092.2024.1042.6056.60
      ShellNet96.3090.4083.9041.0094.2034.7043.9070.20
      GACNet86.4077.7088.5060.6094.2037.3043.5077.80
      SPGraph97.4092.6087.9044.0093.2031.0063.5076.20
      KPConv90.9082.2084.2047.9094.9040.0077.3079.70
      RGNet97.5093.0088.1048.1094.6036.2072.0068.00
      ResDLPS-Net95.6090.7089.2053.4094.7050.8058.9078.60
      SCF-Net97.1091.8086.3051.2095.3050.5067.9080.70
      RandLA-Net95.6091.4086.6051.5095.7051.5069.8076.80
      PVCNN*88.9069.5078.1027.9094.2029.1037.4066.60
      Baseline*97.3090.3087.2047.8094.0044.8057.0079.60
      PVCC-Net*95.6087.5085.4051.5094.3042.8067.6080.90
    • Table 4. Comparison of classification results of different methods on SensatUrban dataset

      View table

      Table 4. Comparison of classification results of different methods on SensatUrban dataset

      MethodIoUs /%
      groundveg.build.wallbridgepark.railtraffic.street.carfoot.bikewater
      PointNet67.9689.5280.050.000.003.950.0031.550.0035.140.000.000.00
      PointNet++72.4694.2484.772.722.0925.790.0031.5411.4238.847.120.0056.93
      SPGraph69.9394.5588.8732.8312.5815.7715.4830.6322.9656.420.540.0044.24
      KPConv87.1098.9195.3374.4028.6941.380.0055.9954.4385.6740.390.0086.30
      BAF-LAC84.4098.4094.1057.2027.6042.5015.0051.6039.5078.1040.100.0075.20
      SCF-Net83.2098.1093.8056.0052.4047.4011.1053.2038.9079.6037.600.0064.70
      RandLA-Net80.1198.0791.5848.8840.7551.620.0056.6733.2380.1432.630.0071.31
      TagentConv71.5491.3875.9035.220.0045.340.0026.6919.2467.580.010.000.00
      SparseConv74.1097.9094.2063.307.5024.200.0030.1034.0074.400.000.0054.80
      PVCNN*80.0097.7092.7034.600.000.000.0040.6010.0067.0013.100.0017.70
      Baseline*82.3097.6093.1056.8040.6046.400.0053.1036.6079.5035.400.0068.70
      PVCC-Net*84.9098.5095.5062.4024.0055.400.0058.7046.8081.3033.800.0079.20
    • Table 5. Classification results of different modules on Toronto3D dataset

      View table

      Table 5. Classification results of different modules on Toronto3D dataset

      MethodVoxelPV-SAPV-CCLovaszOAmIoUIoUs
      roadroad m.naturalbuild.util. l.polecarfence
      A97.7281.0697.5670.7596.7392.4884.2479.6194.1832.95
      B97.9081.4397.9775.7196.3593.7485.5784.0493.9824.10
      C97.9882.3797.9375.8397.1293.4385.7083.9494.0131.01
      D97.9582.6797.8674.9997.0493.6586.7782.0394.6334.38
      PVCC-Net97.9782.9297.8774.9597.0093.9186.6581.7194.8636.42
    • Table 6. Model complexity comparison results

      View table

      Table 6. Model complexity comparison results

      MethodParameters /106FLOPs /106Latency /ms
      RandLA-Net44.9926.38637.39
      SCF-Net1412.1569.25711.84
      BAF-LAC3811.6466.17966.88
      MVPNet39138.47572.201458.29
      PVCNN2215.6165.91724.72
      PVCC-Net49.04206.781133.40
    Tools

    Get Citation

    Copy Citation Text

    Huchen Li, Haiyan Guan, Xiangda Lei, Nannan Qin, Huan Ni. Point‑Voxel Consistency Constraint Network for LiDAR Point Cloud Classification Under Urban Scenes[J]. Chinese Journal of Lasers, 2024, 51(13): 1310003

    Download Citation

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

    Category: remote sensing and sensor

    Received: Nov. 16, 2023

    Accepted: Jan. 4, 2024

    Published Online: Jul. 2, 2024

    The Author Email: Haiyan Guan (guanhy.nj@nuist.edu.cn)

    DOI:10.3788/CJL231411

    CSTR:32183.14.CJL231411

    Topics