• Laser & Optoelectronics Progress
  • Vol. 57, Issue 4, 041510 (2020)
Peng Wang*, Ruizhe Zhu, and Changku Sun
Author Affiliations
  • State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP57.041510 Cite this Article Set citation alerts
    Peng Wang, Ruizhe Zhu, Changku Sun. Point Cloud Coarse Registration Algorithm with Scene Classification Based on Improved RANSAC[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041510 Copy Citation Text show less
    Flow chart of point cloud coarse registration algorithm with scene classification
    Fig. 1. Flow chart of point cloud coarse registration algorithm with scene classification
    Flow chart of scene classification algorithm
    Fig. 2. Flow chart of scene classification algorithm
    Improved RANSAC algorithm
    Fig. 3. Improved RANSAC algorithm
    RGB-D data from fr3_long_office_household. (a) RGB image; (b) depth image
    Fig. 4. RGB-D data from fr3_long_office_household. (a) RGB image; (b) depth image
    RGB-D data from fire. (a) RGB image; (b) depth image
    Fig. 5. RGB-D data from fire. (a) RGB image; (b) depth image
    Point clouds before and after single registration. (a) Before registration; (b) after registration
    Fig. 6. Point clouds before and after single registration. (a) Before registration; (b) after registration
    Point clouds before and after consecutive alignment. (a) Point clouds without alignment; (b) point clouds aligned with coarse registration; (c) point clouds aligned with fine registration; (d) point clouds aligned with truth value
    Fig. 7. Point clouds before and after consecutive alignment. (a) Point clouds without alignment; (b) point clouds aligned with coarse registration; (c) point clouds aligned with fine registration; (d) point clouds aligned with truth value
    General registration errors for different algorithms in each experiment
    Fig. 8. General registration errors for different algorithms in each experiment
    Number of registration failures for different algorithms in each experiment
    Fig. 9. Number of registration failures for different algorithms in each experiment
    LabelPrediction
    -10+1
    -130116
    052710
    +111221
    Table 1. Confusion matrix
    CategoryPRF
    -10.900.760.82
    00.770.820.79
    +10.700.790.74
    Table 2. Evaluation results of SVM
    TypeGeneral registration error /m
    Fr1_deskFr1_roomFr2_deskChessFireHeadsOffice
    Original0.2790.1930.1190.2310.2080.1270.130
    Registered0.0480.0470.0160.0350.0280.0230.020
    Table 3. General registration errors before and after registration
    Peng Wang, Ruizhe Zhu, Changku Sun. Point Cloud Coarse Registration Algorithm with Scene Classification Based on Improved RANSAC[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041510
    Download Citation