Chinese Journal of Lasers, Volume. 52, Issue 1, 0104002(2025)

Dual‐Frequency Virtual‐Stepping Fringe‐Projection Profilometry Driven by Neural Network

Bin Guo1,2, Suodong Ma1,2、*, Junxue Wang1,2, Linxin Liu1,2, Gaonan Miao1,2, and Chinhua Wang1,2
Author Affiliations
  • 1School of Optoelectronic Science and Engineering, Soochow University, Suzhou 215006, Jiangsu , China
  • 2Key Lab of Advanced Optical Manufacturing Technologies of Jiangsu Province & Key Lab of Modern Optical Technologies of Education Ministry of China, Suzhou 215006, Jiangsu , China
  • show less
    Figures & Tables(13)
    Top view of schematic diagram of dual-frequency light-source-stepping method (LSSM)
    Diagrams of the overall system structure and algorithm workflow. (a) Acquisition of dual-frequency single-frame fringe patterns; (b) generation of dual-frequency three-step phase-shifting fringe patterns based on neural networks; (c) 3D reconstruction
    Simulation of LED light source array and planar grating. (a) LED light source array with errors; (b) planar Ronchi grating model
    Simulated high-frequency three-step phase-shifting fringe patterns with errors and the corresponding spectrum. (a1)‒(a3) High-frequency fringe patterns; (a4) normalized intensity curves along the red dashed line in Figs. 4(a1)‒(a3); (b) spectrum of Fig. 4(a1); (c) spectrum cross-sectional line in the center row of Fig. 4(b)
    Absolute phase maps (in radian) of simulated fringe patterns on the reference plane. (a) Absolute phase map obtained by the traditional dual-frequency three-step phase-shifting method; (b) absolute phase map obtained by GVFPS method; (c) phase demodulation error curves along cross-section marked by the red dashed lines
    Three-step phase-shifting fringe deformed patterns output by Res-Unet and their reference ground truth. (a1)‒(a4) Three-step phase-shifting fringe deformed patterns output by the Res-Unet and the normalized intensity curves along the red dashed line; (b1)‒(b4) three-step phase-shifting fringe deformed patterns of the reference ground truth and the normalized intensity curves along the red dashed line
    Absolute phase and the corresponding residual error maps (in radian) of the simulated peak-valley-like object obtained by the traditional dual-frequency three-step phase-shifting method and the proposed method. (a)(c) Traditional dual-frequency three-step phase-shifting method; (b)(d) proposed method; (e) phase demodulation error curves along cross-section marked by the red dashed lines
    Experimental testing device. (a) Overall setup diagram; (b) side view of dual-frequency LSSM projector
    Comparison of fringe patterns for a planar object. (a1)‒(a3) Output of the pre-trained Res-Unet; (a4) normalized intensity curves along the red dashed line; (b1)‒(b3) reference ground truth fringe patterns; (b4) normalized intensity curves along the red dashed line
    Height maps of the planar object obtained by different methods. (a) Height map obtained by the traditional dual-frequency three-step phase-shifting method; (b) height map obtained by the proposed method; (c) comparison curves of height errors along the red dashed lines
    Comparison of fringe patterns for a plaster statue. (a1)‒(a3) Output of the pre-trained Res-Unet; (a4) normalized intensity curves along the red dashed line; (b1)‒(b3) the reference ground truth fringe patterns; (b4) normalized intensity curves along the red dashed line
    Height and corresponding error maps of the plaster statue obtained by different methods. (a)(c) Traditional three-step phase-shifting method; (b)(d) proposed method; (e) comparison curves of height errors along the red dashed lines
    • Table 1. Architecture and parameters of the Res-Unet neural network

      View table

      Table 1. Architecture and parameters of the Res-Unet neural network

      LayeruvsStructure
      EncoderInput1512×512

      Two 3×3 Convolutions+ReLU

      One 2×2 Max Pooling

      132256×256
      264128×128
      312864×64
      425632×32
      Decoder551264×64

      Two 3×3 Convolutions+ReLU

      One 2×2 Up-convolution

      6256128×128

      Copy and Concatenate

      Two 3×3 Convolutions+ReLU

      One 2×2 Up-convolution

      7128256×256
      864512×512
      Output3512×512

      Two 3×3 Convolutions+ReLU

      One 1×1 Convolution

    Tools

    Get Citation

    Copy Citation Text

    Bin Guo, Suodong Ma, Junxue Wang, Linxin Liu, Gaonan Miao, Chinhua Wang. Dual‐Frequency Virtual‐Stepping Fringe‐Projection Profilometry Driven by Neural Network[J]. Chinese Journal of Lasers, 2025, 52(1): 0104002

    Download Citation

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

    Category: Measurement and metrology

    Received: Jun. 24, 2024

    Accepted: Aug. 12, 2024

    Published Online: Jan. 13, 2025

    The Author Email: Ma Suodong (masuodong@suda.edu.cn)

    DOI:10.3788/CJL241001

    CSTR:32183.14.CJL241001

    Topics