Acta Optica Sinica, Volume. 38, Issue 8, 0815017(2018)

Stereo Matching Based on Convolutional Neural Network

Jinsheng Xiao1,2、*, Hong Tian1, Wentao Zou1, Le Tong1, and Junfeng Lei1
Author Affiliations
  • 1 Electronic Information School, Wuhan University, Wuhan, Hubei 430072, China
  • 2 Collaborative Innovation Center of Geospatial Technology, Wuhan, Hubei 430079, China
  • show less
    Figures & Tables(11)
    Matching based on patch. (a) Patch in the left image; (b) patch in the right image
    Structure of MC-CNN net before and after modification
    Loss function curve before and after modification
    Comparison of loss convergence curves before and after network modification
    Disparity map of stereo matching obtained by proposed method (Ⅰ). (a) Original left input image; (b) original right input image; (c) disparity map; (d) ground truth; (e) error graph
    Disparity map of stereo matching obtained by proposed method (Ⅱ). (a) Original left input image; (b) original right input image; (c) disparity map; (d) ground truth; (e) error graph
    • Table 1. Error comparison for the improvement of loss function%

      View table

      Table 1. Error comparison for the improvement of loss function%

      MethodKITTI2012KITTI2015KITTI2012 on KITTI2015KITTI2015 on KITTI2012
      Original loss2.633.284.033.92
      Proposed loss2.613.254.023.89
    • Table 2. Error comparison between different Δ values%

      View table

      Table 2. Error comparison between different Δ values%

      ΔError
      0.023.261
      0.043.266
      0.053.252
      0.063.284
      0.083.267
    • Table 3. [in Chinese]

      View table

      Table 3. [in Chinese]

      Training setKITTI2012KITTI2015
      MC-CNN-slowProposedMC-CNN-slowProposed
      KITTI20122.632.614.014.02
      KITTI20154.323.893.273.25
    • Table 4. Error comparison of disparity with different algorithms (KITTI2012)%

      View table

      Table 4. Error comparison of disparity with different algorithms (KITTI2012)%

      Algorithm>2 pixel>3 pixel>4 pixel>5 pixel
      Elas22.7221.0720.2319.66
      SGM6.284.984.143.57
      SPSS4.863.793.172.76
      Fast CNN4.983.072.392.03
      MC-CNN-fast4.883.032.301.93
      MC-CNN-slow4.28.632.021.72
      Proposed4.362.612.001.70
    • Table 5. Error comparison of disparity with different algorithms (KITTI2015)%

      View table

      Table 5. Error comparison of disparity with different algorithms (KITTI2015)%

      Algorithm>2 pixel>3 pixel>4 pixel>5 pixel
      Elas24.0919.2117.5916.82
      SGM10.036.935.474.48
      SPSS7.154.583.462.93
      Fast CNN6.784.382.562.03
      MC-CNN-fast7.534.012.842.33
      MC-CNN-slow6.383.272.371.97
      Proposed6.563.252.331.92
    Tools

    Get Citation

    Copy Citation Text

    Jinsheng Xiao, Hong Tian, Wentao Zou, Le Tong, Junfeng Lei. Stereo Matching Based on Convolutional Neural Network[J]. Acta Optica Sinica, 2018, 38(8): 0815017

    Download Citation

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

    Category: Machine Vision

    Received: Mar. 27, 2018

    Accepted: May. 11, 2018

    Published Online: Sep. 6, 2018

    The Author Email:

    DOI:10.3788/AOS201838.0815017

    Topics