Laser & Optoelectronics Progress, Volume. 60, Issue 12, 1210009(2023)

Panoramic Three-Dimensional Reconstruction Method Based on Multi-View Encoded Light Field

Zeyu Wang1,2, Sen Xiang1,2、*, Huiping Deng1,2, and Jin Wu1,2
Author Affiliations
  • 1School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China
  • 2Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan 430081, Hubei, China
  • show less
    Figures & Tables(18)
    Framework of proposed method
    Cost curves. (a) Cost curve of proposed method; (b) cost curve of conventional light field
    Schematic diagrams of subpixel resampling. (a) Subpixels at different shear values; (b) subpixel sampling method
    Multi-angle fusion. (a) Sampling schematic; (b) point cloud fusion
    Schematic diagrams of point cloud processing for overlapping parts.(a) Overlapping point clouds; (b) point cloud octree for voxelization; (c) overlapping points; (d) overlapping point processing effect
    Schematics of flying pixel. (a) Generation principle of flying pixels; (b) errors brought by flying pixels; (c) effect after removing flying pixels
    Experimental images. (a) Conventional light field single viewpoint light path; (b) phase-encoded light field single viewpoint light path; (c) 720° multi-view light path, both applicable to conventional light field and phase-encoded light field; (d) 360° horizontal circular light field light path
    Experimental data comparison of dog dataset. (a) MVS; (b) CMVS; (c) conventional light field method[10]; (d) proposed method; (e) ground truth
    Experimental data comparison of head dataset. (a) MVS; (b) CMVS; (c) conventional light field method[10]; (d) proposed method; (e) ground truth
    Experimental data comparison of stone dataset. (a) MVS; (b) CMVS; (c) conventional light field method[10]; (d) proposed method; (e) ground truth
    Results from different angles of proposed method. (a) conventional light field method[10]; (b) proposed method; (c) ground truth
    Comparison of reconstruction effect of dog dataset. (a) 7×7 phase-encoded light field; (b) 5×5 phase-encoded light field; (c) 3×3 phase-encoded light field; (d) 7×7 conventional light field; (e) 5×5 conventional light field; (f) 3×3 conventional light field
    Comparison of reconstruction effect of head dataset. (a) 7×7 phase-encoded light field; (b) 5×5 phase-encoded light field; (c) 3×3 phase-encoded light field; (d) 7×7 conventional light field; (e) 5×5 conventional light field; (f) 3×3 conventional light field
    Comparison of reconstruction effect of stone dataset. (a) 7×7 phase-encoded light field; (b) 5×5 phase-encoded light field; (c) 3×3 phase-encoded light field; (d) 7×7 conventional light field; (e) 5×5 conventional light field; (f) 3×3 conventional light field
    Reconstruction effect of dog dataset under different influencing factors
    • Table 1. Comparison of different datasets

      View table

      Table 1. Comparison of different datasets

      DatasetPanoramic informationNumber of imagesApplicable method
      Conventional light field720°(horizontal 360°+vertical 360°)294Method of reference[10
      Phase-encoded light field720°(horizontal 360°+vertical 360°)294Proposed method
      Circular light fieldhorizontal 360°360CMVS/MVS
    • Table 2. RMSE on different datasets with different angular resolutions

      View table

      Table 2. RMSE on different datasets with different angular resolutions

      RMSE for each datasetdogheadstone
      3×3 encoded0.3090.1820.327
      5×5 encoded0.1230.1020.155
      7×7 encoded0.06690.08150.0838
      3×3 conventional0.6870.7861.272
      5×5 conventional0.3610.4260.602
      7×7 conventional0.1870.2960.450
    • Table 3. RMSE on dog datasets with different influencing factors

      View table

      Table 3. RMSE on dog datasets with different influencing factors

      Influencing

      factor

      Different spatial resolutionRotation angle accuracyPattern with Gaussian noise
      512(proposed)4483840.010.050.1
      RMSE0.06690.1040.1810.1420.1760.2990.5730.8180.1320.1410.157
    Tools

    Get Citation

    Copy Citation Text

    Zeyu Wang, Sen Xiang, Huiping Deng, Jin Wu. Panoramic Three-Dimensional Reconstruction Method Based on Multi-View Encoded Light Field[J]. Laser & Optoelectronics Progress, 2023, 60(12): 1210009

    Download Citation

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

    Category: Image Processing

    Received: Apr. 12, 2022

    Accepted: Jun. 14, 2022

    Published Online: Jun. 5, 2023

    The Author Email: Sen Xiang (xiangsen@wust.edu.cn)

    DOI:10.3788/LOP221282

    Topics