• Laser & Optoelectronics Progress
  • Vol. 58, Issue 24, 2410013 (2021)
Xiaowen Yang*, Aibing Wang, Xie Han, Rong Zhao, and Yuxin Jin
Author Affiliations
  • School of Data Science and Technology, North University of China, Taiyuan, Shanxi 030051, China
  • show less
    DOI: 10.3788/LOP202158.2410013 Cite this Article Set citation alerts
    Xiaowen Yang, Aibing Wang, Xie Han, Rong Zhao, Yuxin Jin. Point Cloud Semantic Segmentation Based on KNN-PointNet[J]. Laser & Optoelectronics Progress, 2021, 58(24): 2410013 Copy Citation Text show less
    Classification effect of points to be measured in local neighborhood
    Fig. 1. Classification effect of points to be measured in local neighborhood
    local feature extraction process
    Fig. 2. local feature extraction process
    KNN-PointNet framework
    Fig. 3. KNN-PointNet framework
    Segmentation results of improved KNN algorithm and KNN algorithm
    Fig. 4. Segmentation results of improved KNN algorithm and KNN algorithm
    Visualization results of scene segmentation under different algorithms. (a) Original point cloud; (b) true segmentation; (c) PointNet; (d) PointNet++
    Fig. 5. Visualization results of scene segmentation under different algorithms. (a) Original point cloud; (b) true segmentation; (c) PointNet; (d) PointNet++
    Visualization results of proposed algorithm
    Fig. 6. Visualization results of proposed algorithm
    Segmentation results under different k and r values
    Fig. 7. Segmentation results under different k and r values
    AlgorithmmAccmIoUoAcc
    PointNet57.841.182.3
    PointNet++--52.4--
    G+RCU--45.1--
    3P-RNN--53.485.7
    Engelmann--52.284.2
    TangentConv--52.882.5
    ASIS--51.1--
    Proposed algorithm81.556.385.8
    Table 1. Performance comparison between improved KNN algorithm and other algorithms unit: %
    TypePointNetPointNet++Proposed network
    Ceiling84.090.590.2
    Floor80.790.091.2
    Wall55.365.274.3
    Beam42.463.370.1
    Pillar26.340.245.2
    Window40.550.356.5
    Door38.642.343.2
    Chair55.068.470.5
    Table30.136.948.3
    Sofa9.632.226.4
    Bookcase27.236.546.3
    Board19.434.238.1
    Clutter25.231.231.6
    Table 2. IoU results of 13 semantic categories on S3DIS dataset for different networks unit: %
    Xiaowen Yang, Aibing Wang, Xie Han, Rong Zhao, Yuxin Jin. Point Cloud Semantic Segmentation Based on KNN-PointNet[J]. Laser & Optoelectronics Progress, 2021, 58(24): 2410013
    Download Citation