Chinese Optics, Volume. 17, Issue 1, 118(2024)

A hybrid network based on light self-limited attention for structured light phase and depth estimation

Xin-jun ZHU*, Hao-miao ZHAO, Hong-yi WANG, Li-mei SONG, and Rui-qun SUN
Author Affiliations
  • School of Artificial Intelligence, Tiangong University, Tianjin 300387, China
  • show less
    Figures & Tables(13)
    Schematic diagram of the FPP system
    Network structure diagram
    Structure of the CNN-Transformer module
    Sample maps in some datasets. The first lines are simulation data, the second lines are real data. (a) Simulation fringe map; (b) simulation fringe map D; (c) simulation fringe map M; (d) simulation fringe wrapped phase; (e) real fringe map; (f) real fringe map D; (g) real fringe map M; (h) real fringe wrapped phase
    Comparison of different network simulation and real data wrapped phases. The blue boxes are the simulation data, and the orange boxes are the real data. (a) UNet; (b) DPH; (c) R2UNet; (d) SUNet; (e) Ours; (f) Label
    Wrapped phase curves.(a) Comparison of simulation data; (b) comparison of real data
    Flowchart of dataset generation. (a) Model import; (b) adjust of the model size; (c) projection fringe
    Sample maps in the dataset. (a) Simulated fringe map; (b) real fringe map; (c) simulation depth map; (d) real depth map
    Comparison of the visual results of depth estimation by different methods. The blue boxes are the simulation data, and the orange boxes are the real data. (a) Input data; (b) UNet; (c) DPH; (d) R2UNet; (e) Ours; (f) Label
    • Table 1. Comparison of the different wrapped phase calculation methods

      View table
      View in Article

      Table 1. Comparison of the different wrapped phase calculation methods

      MSE时间t/s
      直接预测包裹相位0.28335.89
      分别预测DM0.173911.7
      同时预测 DM0.168067.54
    • Table 2. Comparison of the wrapped phase prediction methods

      View table
      View in Article

      Table 2. Comparison of the wrapped phase prediction methods

      仿真数据真实数据
      MSE时间t/sMSE时间t/s
      UNet0.026586.670.168067.54
      DPH0.0271011.650.1297411.78
      R2UNet0.0273413.690.1290514.30
      SUNet0.027177.950.143508.29
      Ours0.0239511.060.1162211.67
    • Table 3. Comparison of ablation experiment results

      View table
      View in Article

      Table 3. Comparison of ablation experiment results

      MSE时间t/s
      CMT11.326.89
      CMT替换LSLA9.176.45
      CMT替换FFN11.345.54
      CMT+U形结构8.949.68
    • Table 4. Comparison of the depth estimation results by different methods

      View table
      View in Article

      Table 4. Comparison of the depth estimation results by different methods

      仿真数据真实数据
      MSE时间t/sMSE时间t/s
      Unet8.785.989.976.44
      DPH8.038.669.8610.59
      R2UNet7.578.738.7210.92
      Ours6.438.097.648.44
    Tools

    Get Citation

    Copy Citation Text

    Xin-jun ZHU, Hao-miao ZHAO, Hong-yi WANG, Li-mei SONG, Rui-qun SUN. A hybrid network based on light self-limited attention for structured light phase and depth estimation[J]. Chinese Optics, 2024, 17(1): 118

    Download Citation

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

    Category:

    Received: Apr. 14, 2023

    Accepted: --

    Published Online: Mar. 28, 2024

    The Author Email:

    DOI:10.37188/CO.2023-0066

    Topics