Laser & Optoelectronics Progress, Volume. 62, Issue 12, 1215004(2025)

Point Cloud Registration Method Based on Three-Dimensional Linear Multi-Feature Description

Zhengpeng Zhang1, Jianhua Liu1、*, Xinyu Xie2, Lijing Bu1, and Yan Cheng1
Author Affiliations
  • 1College of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, Hunan , China
  • 2Great Wall Information Co., Ltd., Changsha 410199, Hunan , China
  • show less
    Figures & Tables(12)
    Multi-scale curvature points
    Three-dimensional straight line fitting and extraction. (a) First fit; (b) eliminate outliers; (c) second fitting
    Multi-feature description of 3D straight lines. (a) Straight line of source point cloud; (a') straight line of target point cloud; (b)(b') angular features; (c)(c') line-line distance feature; (d)(d') point-line distance feature
    Three-dimensional straight line extraction results
    Comparative test of point cloud registration
    Point cloud registration at extreme locations . (a) Upside down 1; (b) upside down 2; (c) sideways; (d) symmetry
    • Table 1. Straight line fitting parameters

      View table

      Table 1. Straight line fitting parameters

      DateNumber of pointsFitting number of key pointsFirst extraction number of pointsSecond extraction number of pointsNumber of line
      Date 135947147890627323
      Date 2434671777130636923
      Date 3780563274257588030
      Date 4484852151153741029
      Date 5200000100178817163270
      Date 62000001001451711295110
      Date 729915119778425017
      Date 8200000100078453178068
    • Table 2. Effect of key point extraction ratio on the registration results

      View table

      Table 2. Effect of key point extraction ratio on the registration results

      DataRegistration results at extraction ratio of 0.01Registration results at extraction ratio of 0.05Registration results at extraction ratio of 0.1
      RMSE /(10-2 m)Time /sPoint quantityRMSE /(10-2 m)Time /sPoint quantityRMSE /(10-2 m)Time /sPoint quantity
      Data 10.1050.333580.0500.7817960.0181.253593
      Data 20.0880.304330.0620.6621720.0280.624345
      Data 30.1010.347790.0700.5139020.0240.637804
      Data 40.1000.334830.0820.5924250.0180.714847
      Data 50.0660.4420030.0174.52100000.00214.7320006
      Data 60.4090.4920020.070140.41999820000
      Data 70.4980.302980.0120.4914900.0180.522984
      Data 80.0650.3120050.0282.47100000.0044.7120010
    • Table 3. Experimental results of point cloud registration

      View table

      Table 3. Experimental results of point cloud registration

      DateAlgorithmFS /(10-5 m)RMSE /(10-2 m)Time /sDateAlgorithmFS /(10-5 m)RMSE /(10-2 m)Time /s
      Date 13DSC9.2480.31419.38Date 53DSC4.4140.20224.62
      4PCS3.6680.2611.484PCS7.3670.6432.72
      FPFH2.3670.1924.71FPFH1.6090.15511.46
      PFH3.5980.24031.42PFH1.6210.13055.03
      RANSAC2.9890.2104.29RANSAC1.9350.07939.08
      SHOT1.7160.1647.19SHOT1.7720.16714.90
      Ours0.0260.0500.78Ours0.0030.0174.28
      Date 23DSC4.5770.24321.22Date 63DSC599.1040.78918.39
      4PCS7.0000.2443.074PCS2424.6103.2095.43
      FPFH9.3150.2043.62FPFH1.4620.168183.61
      PFH14.7230.23913.97PFH1.6060.20289.68
      RANSAC5.1650.1512.02RANSAC1.9230.079440.19
      SHOT7.3000.1584.89SHOT2.2660.299119.94
      Ours0.0450.0620.66Ours0.0520.070140.41
      Date 33DSC0.2460.06028.52Date 73DSC1.6110.15223.69
      4PCS5.5740.5623.644PCS3.6530.4980.64
      FPFH1.0360.0563.89FPFH1.8960.1852.67
      PFH1.2930.0809.70PFH1.9210.1126.86
      RANSAC1.5740.0937.66RANSAC1.4160.0549.52
      SHOT1.8310.1054.97SHOT7.7040.6568.88
      Ours0.0920.0700.51Ours0.0680.0120.49
      Date 43DSC0.6410.14121.55Date 83DSC6.7830.11630.33
      4PCS8.1980.6890.954PCS5.2350.4861.48
      FPFH1.7110.1923.86FPFH1.9450.0818.90
      PFH2.1470.22920.68PFH1.8010.08627.12
      RANSAC1.8340.11410.52RANSAC2.0330.0849.74
      SHOT1.8050.1845.35SHOT1.4480.10710.23
      Ours0.0810.0820.59Ours0.0100.0282.47
    • Table 4. Fine registration experimental results of different methods with ICP

      View table

      Table 4. Fine registration experimental results of different methods with ICP

      DataRMSE /m
      3DSC+ICP4PCS+ICPFPFH+ICPPFH+ICPRANSAC+ICPSHOT+ICPOurs
      Date 15.87×10-61.867×10-60.0007901.855×10-83.504×10-82.003×10-81.081×10-8
      Date 20.0003820.0002930.0003820.0005530.0003390.0002656.552×10-7
      Date 30.0002580.0004900.0002350.0002580.0002350.0002354.388×10-7
      Date 48.592×10-97.832×10-91.395×10-82.415×10-89.357×10-91.565×10-83.229×10-6
      Date 50.001280.0006550.0006550.0006490.0006530.0006532.048×10-8
      Date 60.003520.003490.003470.003470.003480.003475.340×10-7
      Date 71.455×10-82.061×10-62.203×10-81.449×10-81.396×10-81.549×10-84.043×10-7
      Date 80.0004320.0004320.0004330.0004350.0004320.0004332.471×10-8
    • Table 5. Compared with the improved point registration methods

      View table

      Table 5. Compared with the improved point registration methods

      DateComparison methodRMSE /(10-2 m)Time /s
      Date 1Method 173.751
      Method 286.299
      Method 392.5003.893
      Method 4120.0910.376
      Method 5130.0301.530
      Method 6212.1474.384
      Method 7220.08011.930
      Method 8230.3604.692
      Ours0.0500.281
    • Table 6. Point cloud registration results at extreme locations

      View table

      Table 6. Point cloud registration results at extreme locations

      Extreme locationRMSE /(10-2 m)
      3DSC4PCSFPFHPFHRANSACSHOTOurs
      Upside down, see Fig.6 (a)0.34171.13110.38740.34810.13410.20780.0186
      Upside down, see Fig.6 (b)0.12110.74900.28710.25260.14600.16000.0209
      Sideways, see Fig.6(c)0.29310.95520.29930.26350.18990.22410.0472
      Symmetry, see Fig.6(d)0.29370.73800.16750.23180.18130.34510.0153
    Tools

    Get Citation

    Copy Citation Text

    Zhengpeng Zhang, Jianhua Liu, Xinyu Xie, Lijing Bu, Yan Cheng. Point Cloud Registration Method Based on Three-Dimensional Linear Multi-Feature Description[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1215004

    Download Citation

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

    Category: Machine Vision

    Received: Nov. 28, 2024

    Accepted: Dec. 12, 2024

    Published Online: Jun. 10, 2025

    The Author Email: Jianhua Liu (202221623122@smail.xtu.edu.cn)

    DOI:10.3788/LOP242345

    CSTR:32186.14.LOP242345

    Topics