• Laser & Optoelectronics Progress
  • Vol. 58, Issue 12, 1210001 (2021)
Wanting Song, Wensong Jiang*, and Zai Luo**
Author Affiliations
  • College of Metrology and Measurement Engineering, China Jiliang University, Hangzhou, Zhejiang 310018, China
  • show less
    DOI: 10.3788/LOP202158.1210001 Cite this Article Set citation alerts
    Wanting Song, Wensong Jiang, Zai Luo. Rapid Batch Three-Dimensional Reconstruction of Point Clouds Based on Multi-Label Classification[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210001 Copy Citation Text show less
    Preprocessing flow chart of 3D point clouds for neural network
    Fig. 1. Preprocessing flow chart of 3D point clouds for neural network
    Schematic diagram of voxel grid algorithm
    Fig. 2. Schematic diagram of voxel grid algorithm
    Subject point cloud and noise point cloud
    Fig. 3. Subject point cloud and noise point cloud
    Model architecture for point clouds classification
    Fig. 4. Model architecture for point clouds classification
    Delaunay triangulation. (a) Insert point; (b) judgement of the common inner edge using empty circumcircle; (c) delete the inner edge; (d) insert completed
    Fig. 5. Delaunay triangulation. (a) Insert point; (b) judgement of the common inner edge using empty circumcircle; (c) delete the inner edge; (d) insert completed
    Overall flow chart
    Fig. 6. Overall flow chart
    Example diagrams of 3D models. (a) Sphere; (b) cube; (c) cylinder; (d) penetrant; (e) open-source table point cloud set; (f) open-source chair point cloud set
    Fig. 7. Example diagrams of 3D models. (a) Sphere; (b) cube; (c) cylinder; (d) penetrant; (e) open-source table point cloud set; (f) open-source chair point cloud set
    Comparison of denoising effect of the standard ball. (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Fig. 8. Comparison of denoising effect of the standard ball. (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Comparison of denoising effect of the standard gauge block: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Fig. 9. Comparison of denoising effect of the standard gauge block: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Comparison of denoising effect of the standard cylinder: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Fig. 10. Comparison of denoising effect of the standard cylinder: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Comparison of denoising effect of the open-source table point cloud set: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Fig. 11. Comparison of denoising effect of the open-source table point cloud set: (a) k-neighbor denoising algorithm; (b) radius filtering algorithm; (c) proposed denoising algorithm
    Processing flow
    Fig. 12. Processing flow
    Relationship between reconstruction number and reconstruction time when the number of point clouds is 100000
    Fig. 13. Relationship between reconstruction number and reconstruction time when the number of point clouds is 100000
    Relationship between reconstruction number and reconstruction time when the number of point clouds is 700000
    Fig. 14. Relationship between reconstruction number and reconstruction time when the number of point clouds is 700000
    Relationship between reconstruction number and reconstruction time when the number of point clouds is 1200000
    Fig. 15. Relationship between reconstruction number and reconstruction time when the number of point clouds is 1200000
    Point cloud setk-neighbor denoising algorithmRadius filtering algorithmProposed denoising algorithm
    Standard ball10364210839288349
    Standard gauge block671400687593659876
    Standard cylinder112516311498931116856
    Open-source table point cloud set451306452099429705
    Table 1. Number of point clouds after denoising by the three methods
    Number of initial point cloudsTraditional preprocessing methodProposed preprocessing method
    Denoising time ta1/sSimplifying time ta2/sTotal time ta/sSimplifying time tb1/sDenoising time tb2/sTotal time tb/s
    1089570.9730.0250.9980.0530.0260.079
    7289071.8800.0671.9470.0940.1340.228
    12630849.4600.2389.6980.3060.5890.895
    Table 2. Comparison of the time required for point cloud preprocessing
    Number of initial point cloudsTraditional preprocessing methodProposed preprocessing method
    Number of point clouds after denoisingNumber of point clouds after simplifyingNumber of point clouds after simplifyingNumber of point clouds after denoising
    108957103642106315211346
    728907671400478961135897
    12630841125163120402096320738
    Table 3. Comparison of point cloud number after preprocessing
    Test No.Test accuracy a1Test accuracy b1
    10.9040.907
    20.9010.905
    30.9130.921
    40.9090.904
    50.9170.905
    Table 4. Comparison of accuracy
    Number of 3Dpoint cloudsTime required by the traditional method t/sTime required by the proposed method t'/s
    1010.080.87
    2021.782.18
    4041.543.92
    6062.125.98
    8082.467.64
    100101.949.54
    120122.0811.98
    Table 5. Comparison of reconstruction time when the number of point clouds is 100000
    Number of 3D point cloudsTime required by the traditional method t/sTime required by the proposed method t'/s
    1020.332.36
    2042.084.78
    4082.749.52
    60123.5614.34
    80164.1419.26
    100204.0623.78
    120245.2829.06
    Table 6. Comparison of reconstruction time when the number of point clouds is 700000
    Number of 3Dpoint cloudsTime required by thetraditional method t/sTime required by the proposed method t'/s
    1097.979.03
    20194.5917.94
    40393.7237.06
    60588.5456.49
    80780.9771.96
    100980.5692.44
    1201178.41109.98
    Table 7. Comparison of reconstruction time when the number of point clouds is 1200000
    Wanting Song, Wensong Jiang, Zai Luo. Rapid Batch Three-Dimensional Reconstruction of Point Clouds Based on Multi-Label Classification[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210001
    Download Citation