• Laser & Optoelectronics Progress
  • Vol. 59, Issue 16, 1615006 (2022)
Haohao Zhou, Xiaoxu Wang, Jinglong Wang, and Kangsheng Lai*
Author Affiliations
  • School of Optoelectronic Engineering and Instrumentation Science, Dalian University of Technology, Dalian 116024, Liaoning , China
  • show less
    DOI: 10.3788/LOP202259.1615006 Cite this Article Set citation alerts
    Haohao Zhou, Xiaoxu Wang, Jinglong Wang, Kangsheng Lai. Matching Algorithm Based on Improved Cost Calculation and Path Optimization Strategy[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1615006 Copy Citation Text show less
    Diagram of proposed method
    Fig. 1. Diagram of proposed method
    Process of matching cost calculation
    Fig. 2. Process of matching cost calculation
    Transform effects of improved LBP operator and initial LBP operator. (a) Source image; (b) result of initial LBP operator; (c) result of improved LBP operator; (d) calculation time comparison
    Fig. 3. Transform effects of improved LBP operator and initial LBP operator. (a) Source image; (b) result of initial LBP operator; (c) result of improved LBP operator; (d) calculation time comparison
    Traditional eight-path cost aggregation
    Fig. 4. Traditional eight-path cost aggregation
    Path selection in proposed method
    Fig. 5. Path selection in proposed method
    Experimental effect images. (a) Raw images in left view (tsukuba, venus, teddy, and cones); (b) true disparity map; (c) disparity map of SGM; (d) disparity map of proposed algorithm; (e) disparity map in pseudo-color of proposed algorithm
    Fig. 6. Experimental effect images. (a) Raw images in left view (tsukuba, venus, teddy, and cones); (b) true disparity map; (c) disparity map of SGM; (d) disparity map of proposed algorithm; (e) disparity map in pseudo-color of proposed algorithm
    ParameterLBP sizeP1P2Rξζλ
    Value7×71015051025
    Table 1. Parameters of proposed algorithm
    AlgorithmTsukubaVenusTeddyCones
    d∈[0,16]d∈[0,20]d∈[0,64]d∈[0,64]
    Cost calculationCost aggregationCost calculationCost aggregationCost calculationCost aggregationCost calculationCost aggregation
    SGM0.1708.4720.28015.7770.69549.9870.71350.114
    Proposed0.0535.3540.07410.0760.25329.9790.25429.714
    Improved time

    0.117

    (68.0%

    3.118

    (36.8%

    0.206

    (73.5%

    5.701

    (36.1%

    0.442

    (63.5%

    20.008

    (40.0%

    0.459

    (64.3%

    20.400

    (40.7%

    Table 2. Calculation time comparison in debug mode
    AlgorithmTsukubaVenusTeddyCones
    d∈[0,16]d∈[0,20]d∈[0,64]d∈[0,64]
    Cost calculationCost aggregationCost calculationCost aggregationCost calculationCost aggregationCost calculationCost aggregation
    SGM0.0370.0880.0710.1630.1960.4450.2050.493
    Proposed0.0170.0450.0250.0970.0750.3110.0930.289
    Improved time

    0.020

    (54.0%

    0.043

    (48.8%

    0.046

    (64.7%

    0.066

    (40.5%

    0.121

    (61.7%

    0.134

    (30.1%

    0.112

    (54.6%

    0.204

    (41.3%

    Table 3. Calculation time comparison in release mode
    AlgorithmTsukubaVenusTeddyConesAverage
    nonoccalldiscnonoccalldiscnonoccalldiscnonoccalldisc
    RTcensus195.16.319.21.62.414.28.013.820.34.19.512.29.73
    ADcensus201.11.55.70.10.31.24.16.210.92.47.37.03.97
    C-SemiGlob232.63.39.90.30.63.25.111.813.02.88.48.25.76
    planeFitSGM213.14.214.91.11.914.65.711.617.13.89.311.38.21
    SGM143.34.012.81.01.611.36.012.216.33.19.88.97.51
    SGMDDW222.34.411.81.22.716.86.514.517.55.614.214.89.36
    Proposed3.33.512.70.91.19.87.39.114.43.97.311.77.08
    Table 4. Comparison of mismatched ratio with threshold is 1
    AlgorithmTsukubaVenusTeddyConesAverage
    nonoccalldiscnonoccalldiscnonoccalldiscnonoccalldisc
    RTcensus1912.914.128.13.74.617.811.418.627.75.511.815.914.34
    ADcensus2026.827.021.14.14.68.010.613.820.16.612.411.913.92
    C-SemiGlob2313.914.718.93.33.810.99.817.422.85.411.712.812.12
    planeFitSGM219.210.423.32.33.215.99.017.025.55.211.614.712.28
    SGM1413.414.320.34.65.415.711.018.526.14.912.513.513.35
    SGMDDW2215.517.523.18.19.625.615.723.931.514.522.826.019.48
    Proposed6.38.920.15.55.917.38.210.030.58.510.117.412.39
    Table 5. Comparison of mismatched ratio when threshold is 0.5

    Stereo

    pairs

    Mismatched ratio
    MDP23SRM24Cens525SGBM113SGM14DTS26Proposed
    Adiron25.028.737.039.529.120.925.5
    ArtL15.115.817.519.011.517.711.7
    Jadepl35.938.235.936.028.134.320.2
    Motor30.427.632.525.925.521.117.9
    MotorE28.730.328.936.922.521.815.8
    Piano33.638.133.636.626.133.022.3
    PianoL41.747.947.458.642.139.039.5
    Pipes29.023.223.620.420.524.914.4
    Playrm42.744.647.445.038.340.626.8
    Playt47.835.067.952.464.734.745.3
    PlaytP28.832.929.129.224.732.217.3
    Recyc31.332.433.633.027.024.118.9
    Shelvs54.554.462.157.259.046.941.3
    Teddy10.810.612.318.110.110.38.2
    Vintge43.649.561.255.051.347.635.9
    Austr36.329.363.251.858.124.540.7
    AustrP18.427.219.024.110.223.617.1
    Bicyc224.425.619.221.415.517.517.9
    Class30.029.533.724.430.421.731.3
    ClassE48.840.857.378.554.638.338.2
    Compu16.918.930.020.821.220.917.9
    Crusa40.745.561.347.848.844.334.2
    CrusaP32.845.247.741.927.945.019.5
    Djemb20.921.819.213.413.919.223.7
    DjembL47.342.549.763.544.133.830.9
    Hoops41.241.551.249.844.239.331.9
    Livgrm28.833.439.534.633.129.123.2
    Nkuba38.338.736.434.632.436.120.7
    Plants33.532.343.035.431.931.722.3
    Stairs35.242.555.059.244.729.831.3
    Average30.932.436.635.229.728.625.4
    Table 6. Comparison of mismatched ratio in all regions
    Haohao Zhou, Xiaoxu Wang, Jinglong Wang, Kangsheng Lai. Matching Algorithm Based on Improved Cost Calculation and Path Optimization Strategy[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1615006
    Download Citation