Acta Optica Sinica, Volume. 44, Issue 11, 1112003(2024)

High-Precision Visual SLAM Method Based on Industrial Reflective Features

Zhao Guo, Ze Yang, Yongjie Ren, Yanbiao Sun*, and Jigui Zhu
Author Affiliations
  • School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China
  • show less
    Figures & Tables(22)
    Framework of proposed visual SLAM system
    Composition of coding features and identification of coding value
    Simultaneous localization and mapping based on sliding window
    Normal vector solution of coding features
    Schematic of camera clustering
    Selection of key camera views (selected key frame is C2, and the corresponding red area is deleted)
    Apply global pose constraints to current frame (Fi is the current image frame, and the remaining three frames are global key frames)
    Experimental site
    Experimental equipment
    Global map construction result
    Ground truth and trajectory of ORB-SLAM3 algorithm obtained by natural and reflective features respectively
    Three-dimensional view of four groups of tracks
    Ground truth and four motion trajectories obtained by proposed method, PnP, and ORB-SLAM3. (a) Sequence 1; (b) sequence 2; (c) sequence 3; (d) sequence 4
    Absolute trajectory error distribution of four groups of data. (a) Sequence 1; (b) sequence 2; (c) sequence 3; (d) sequence 4
    Attitude change curves of four groups of data. (a)-(c) Rotation of sequence 1 around three axis; (d)-(f) rotation of sequence 2 around three axis; (g)-(i) rotation of sequence 3 around three axis; (j)-(l) rotation of sequence 4 around three axis
    Relative attitude error curves of four groups of data. (a)-(c) Rotation of sequence 1 around three axis; (d)-(f) rotation of sequence 2 around three axis; (g)-(i) rotation of sequence 3 around three axis; (j)-(l) rotation of sequence 4 around three axis
    • Table 1. Basic parameters of main equipments

      View table

      Table 1. Basic parameters of main equipments

      EquipmentParameter
      Leica M10

      Resolution: 7840 pixel×5184 pixel

      Focal length: 21 mm

      Pixel size: 4.59 μm

      IDS UI-5280SE-M-GL

      Resolution: 2048 pixel×1536 pixel

      Focal length: 8 mm

      Pixel size: 3.45 μm

      Xsens MTi-100 IMU

      Gyroscopes noise density: 0.01 (°)/(s·Hz)

      Accelerometers noise density: 60 μg/Hz

    • Table 2. 3D position measurement accuracy of map points (comparison reference: global BA results of all images)

      View table

      Table 2. 3D position measurement accuracy of map points (comparison reference: global BA results of all images)

      DirectionIND methodProposed method
      RMS error /mmMax error /mmRMS error /mmMax error /mm
      X direction0.573.520.311.25
      Y direction0.180.630.170.48
      Z direction0.341.660.380.99
      Absolute trajectory0.693.660.521.29
    • Table 3. Number of frames involved in the BA and the number and proportion of bad points with errors greater than 1 mm

      View table

      Table 3. Number of frames involved in the BA and the number and proportion of bad points with errors greater than 1 mm

      MethodNumber of key framesNumber of points (error >1 mm)Proportion of points (error >1 mm) /%
      IND31347.69
      Proposed57132.95
    • Table 4. Absolute trajectory error (ATE ) in global localization (comparison reference: T-Mac)

      View table

      Table 4. Absolute trajectory error (ATE ) in global localization (comparison reference: T-Mac)

      ATE

      ORB-SLAM3

      (natural points)

      ORB-SLAM3

      (reflective points)

      ATE of X direction/ mm51.2811.50
      ATE of Y direction / mm37.336.36
      ATE of Z direction / mm34.7612.85
      Total ATE / mm72.2618.42
    • Table 5. ATE and three-axis displacement error in global localization (comparison reference: T-Mac)

      View table

      Table 5. ATE and three-axis displacement error in global localization (comparison reference: T-Mac)

      Sequence1234
      RMSE of X displacement /mmORB-SLAM3 (reflective points)1.633.932.027.46
      PnP0.761.231.151.26
      Proposed0.740.951.011.13
      RMSE of Y displacement /mmORB-SLAM3 (reflective points)0.952.901.467.73
      PnP0.331.180.711.22
      Proposed0.340.930.780.98
      RMSE of Z displacement /mmORB-SLAM3 (reflective points)1.201.032.433.19
      PnP0.401.380.851.59
      Proposed0.340.980.791.30
      RMSE of ATE /mORB-SLAM3 (reflective points)2.244.993.4811.21
      PnP0.922.201.602.36
      Proposed0.881.651.501.98
    • Table 6. Relative attitude error in global localization (comparison reference: T-Mac)

      View table

      Table 6. Relative attitude error in global localization (comparison reference: T-Mac)

      Sequence

      RMSE of

      ORB-SLAM3

      (reflective points) /(°)

      RMSE of PnP /(°)RMSE of proposed method /(°)
      10.0230.0120.015
      20.0350.0510.023
      30.0330.0270.030
      40.2400.0790.024
    Tools

    Get Citation

    Copy Citation Text

    Zhao Guo, Ze Yang, Yongjie Ren, Yanbiao Sun, Jigui Zhu. High-Precision Visual SLAM Method Based on Industrial Reflective Features[J]. Acta Optica Sinica, 2024, 44(11): 1112003

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Feb. 5, 2024

    Accepted: Mar. 15, 2024

    Published Online: Jun. 17, 2024

    The Author Email: Sun Yanbiao (yanbiao.sun@tju.edu.cn)

    DOI:10.3788/AOS240611

    CSTR:32393.14.AOS240611

    Topics