Acta Photonica Sinica, Volume. 53, Issue 8, 0810003(2024)

Dynamic Weight Cost Aggregation Algorithm for Stereo Matching Based on Adaptive Window

Fupei WU... Yuhao LIU, Rui WANG and Shengping LI* |Show fewer author(s)
Author Affiliations
  • Department of Mechanical Engineering, College of Engineering, Shantou University, Shantou 515063, China
  • show less
    Figures & Tables(21)
    The adaptive cross domain
    The construction of adaptive window diagram
    The two-pass cost aggregation diagram
    The neighborhood window of invalid pixel
    Result of SLIC algorithm
    Influence of α and β on error matching rate of disparity images
    Results of disparity optimization
    Experimental results
    Experimental results of proposed algorithm on Middlebury image pairs
    The acquired image and the experimental platform
    The flowchart of image processing
    Comparisons of three-dimensional reconstruction results
    The reconstruction results of four surface types using different algorithms
    • Table 1. Experimental parameter settings

      View table
      View in Article

      Table 1. Experimental parameter settings

      ParameterValueParameterValue
      λAD10λCensus30
      τ115a100
      τ226δ20
      L150α4
      L225β11
      τS20τH0.4
    • Table 2. Average error matching rate in non-occlusion region and all region(%)

      View table
      View in Article

      Table 2. Average error matching rate in non-occlusion region and all region(%)

      AlgorithmTeddyBaby1Cloth3Wood2Average
      n-occalln-occalln-occalln-occall
      Mei's aggregation5.918.846.547.583.183.966.887.936.35
      Proposed algorithm4.937.084.786.882.633.011.832.544.21
    • Table 3. Error matching rates of different algorithms in non-occlusion region and all region (%)

      View table
      View in Article

      Table 3. Error matching rates of different algorithms in non-occlusion region and all region (%)

      AlgorithmTeddyBaby1Baby2Baby3FlowerpotsAverage
      n-occalln-occalln-occalln-occalln-occalln-occall
      LSECVR4.9116.085.128.806.619.797.6114.4814.6619.837.7813.80
      GF5.779.626.539.787.9710.535.708.5810.6414.417.3210.58
      Semiglob6.2512.367.3610.989.4114.987.1310.3312.7717.348.5813.20
      AD-Census4.937.084.786.887.338.136.858.979.1212.886.608.79
      Patchmatch3.565.035.386.536.237.866.447.018.5610.686.037.42
      Proposed algorithm2.523.163.864.984.126.083.435.266.638.794.125.65
    • Table 4. Execute time of six algorithms (s)

      View table
      View in Article

      Table 4. Execute time of six algorithms (s)

      AlgorithmLSECVRGFSemiglobAD-CensusPatchmatchProposed
      Execute time/s1.571.960.981.5630.842.53
    • Table 5. Camera parameters

      View table
      View in Article

      Table 5. Camera parameters

      Parameter matrixLeft cameraRight camera
      Inner parameter matrix17910632.801791453.500117920643.301792512.6001
      Distortion coefficient-0.140-0.135002.511-0.136-0.389005.813
      Rotation matrix0.981 5-0.015 40.190 90.014 60.999 90.005 6-0.191 0-0.002 70.981 6
      Translation vector-48.20-0.78811.26
    • Table 6. Experimental results of three-dimensional reconstruction

      View table
      View in Article

      Table 6. Experimental results of three-dimensional reconstruction

      Convex surfaceStep surfaceAngular surfaceConcave surface
      Sampleimage
      Acquireimage
      Reconstructionimage
    • Table 7. Error measuring rates of sample in length

      View table
      View in Article

      Table 7. Error measuring rates of sample in length

      NameActual length/mmMeasure length/mmError/mmError rate/%
      Convex surface1615.873 2-0.126 8-0.793
      87.974 4-0.025 6-0.320
      Step surface2525.299 80.299 81.199
      109.954 3-0.045 7-0.457
      Angular surface1616.157 50.157 50.984
      88.047 70.047 70.596
      Concave surface2019.836 9-0.163 1-0.815
      109.962 4-0.037 6-0.376
    • Table 8. Error measuring rates of sample in height

      View table
      View in Article

      Table 8. Error measuring rates of sample in height

      NameMaximum error/mmAverage error/mmAverage absolute error/mmSD error/mm
      Convex surface0.233 40.081 20.112 30.098 1
      Step surface0.144 8-0.100 40.061 30.066 1
      Angular surface0.221 7-0.007 40.036 80.051 8
      Concave surface0.336 3-0.054 10.103 50.091 7
      Ave0.234 1-0.020 20.078 50.076 9
    Tools

    Get Citation

    Copy Citation Text

    Fupei WU, Yuhao LIU, Rui WANG, Shengping LI. Dynamic Weight Cost Aggregation Algorithm for Stereo Matching Based on Adaptive Window[J]. Acta Photonica Sinica, 2024, 53(8): 0810003

    Download Citation

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

    Category:

    Received: Jan. 24, 2024

    Accepted: Mar. 21, 2024

    Published Online: Oct. 15, 2024

    The Author Email: LI Shengping (spli@stu.edu.cn)

    DOI:10.3788/gzxb20245308.0810003

    Topics