光学学报, 2020, 40 (9): 0915001, 网络出版: 2020-05-06   

基于引导图像和自适应支持域的立体匹配 下载: 1240次

Stereo Matching Based on Guidance Image and Adaptive Support Region
作者单位
四川大学计算机学院, 四川 成都 610065
图 & 表

图 1. 本文算法流程图

Fig. 1. Flowchart of proposed algorithm

下载图片 查看原文

图 2. 十字交叉法构建自适应支持域示意图

Fig. 2. Schematic of constructing adaptive support region based on cross method

下载图片 查看原文

图 3. 三种代价计算方法的视差图。(a) C1;(b) C2;(c)所提梯度计算方法

Fig. 3. Disparity maps of three cost computation methods. (a) C1; (b) C2; (c) proposed gradient calculation method

下载图片 查看原文

图 4. 全部区域和非遮挡区域的加权平均值。(a) Avgerr;(b) RMSE

Fig. 4. Weighted averages for all regions and non-occluded regions. (a) Avgerr; (b) RMSE

下载图片 查看原文

图 5. 每个环节进行视差细化后的加权平均值。(a)(b) Avgerr; (c)(d) RMSE

Fig. 5. Weighted average after disparity refinement on each step. (a)(b) Avgerr; (c)(d) RMSE

下载图片 查看原文

图 6. 视差结果对比。(a) Adirondack;(b) Jadeplant;(c) Piano;(d) Motorcycle;(e) Recycle

Fig. 6. Comparison of disparity results. (a) Adirondack; (b) Jadeplant; (c) Piano; (d) Motorcycle; (e) Recycle

下载图片 查看原文

表 1视差细化前后误差的加权平均值和降低的百分比(所提梯度计算方法的视差图)

Table1. Weighted average of errors before and after disparity refinement and reduced percentage (disparity map of proposed gradient calculation method)

ParameterWeighted average /pixelReducedpercentage /%
Before disparity refinementAfter disparity refinement
Avgerr(all)21.111.3046.4
Avgerr(nonocc)12.17.8135.5
RMSE(all)47.227.7041.3
RMSE(nonocc)31.620.9033.9

查看原文

表 2视差细化前后误差的加权平均值和降低的百分比(C1的视差图)

Table2. Weighted average of errors before and after disparity refinement and reduced percentage (disparity map of C1)

ParameterWeighted average /pixelReduced percentage /%
Before disparity refinementAfter disparity refinement
Avgerr(all)22.812.4045.6
Avgerr(nonocc)13.68.6336.5
RMSE(all)49.629.8039.9
RMSE(nonocc)34.522.9033.6

查看原文

表 3视差细化前后误差的加权平均值和降低的百分比(C2的视差图)

Table3. Weighted average of errors before and after disparity refinement and reduced percentage (disparity map of C2)

ParameterWeighted average /pixelReduced percentage /%
Before disparity refinementAfter disparity refinement
Avgerr(all)24.514.939.2
Avgerr(nonocc)15.110.729.1
RMSE(all)51.534.632.8
RMSE(nonocc)36.327.125.3

查看原文

表 4全部区域的Avgerr对比

Table4. Comparison of Avgerr in all regionspixel

Image nameLE-ELASIEBIMstSPSSM-AWPDSGCADoGGuidedProposed algorithm
Adirondack9.3127.306.5110.507.6820.106.40
ArtL5.9015.1015.2019.9021.7028.009.00
Jadeplant64.5055.6040.0062.7045.0056.5026.10
Motorcycle7.245.548.3511.0010.6013.808.11
MotorcycleE7.658.218.4512.5010.4016.8011.40
Piano6.256.4012.009.0811.5013.406.15
PianoL9.6918.9025.0029.7024.5037.3034.00
Pipes12.8011.8016.1021.1019.9023.8014.90
Playroom10.1018.0025.2020.7024.6030.3010.50
Playtable23.9017.9015.709.5034.5030.8016.70
PlaytableP4.274.9512.409.7514.8013.0010.00
Recycle7.395.298.817.187.569.134.20
Shelves8.4817.1023.7011.4017.3019.009.97
Teddy2.985.318.019.4412.2013.403.35
Vintage14.0010.9053.7016.8043.8013.6010.90
Australia15.209.178.6419.1016.6018.2012.00
AustraliaP6.945.548.7718.2012.4012.608.31
Bicycle26.687.5411.4016.0012.9017.6013.70
Classroom224.6027.9020.2029.3032.6034.909.09
Classroom2E69.6055.0027.0051.1039.3076.3067.10
Computer12.4013.8022.2022.5020.6022.1013.20
Crusade21.7074.3050.8091.8049.5073.4036.30
CrusadeP21.0074.6050.2094.9050.5071.3035.60
Djembe2.732.103.657.335.716.642.98
DjembeL13.8029.1017.2031.8024.5039.0019.50
Hoops22.8045.0038.7037.7036.3056.6023.00
Livingroom10.309.4930.4016.8022.9025.907.18
Newkuba16.2013.3020.3028.5023.2028.7011.30
Plants43.3023.3026.2032.2027.7033.9025.80
Staircase21.3030.9039.4036.4039.8057.5029.80
Weighted average15.9520.9520.7026.5022.8029.2015.20

查看原文

表 5非遮挡区域的Avgerr对比

Table5. Comparison of Avgerr in non-occluded regionspixel

Image nameLE-ELASIEBIMstSPSSM-AWPDSGCADoGGuidedProposed algorithm
Adirondack8.4626.103.576.313.2515.204.84
ArtL3.834.675.349.655.959.574.62
Jadeplant41.1041.9022.8031.8018.9027.1016.10
Motorcycle5.122.723.114.713.605.644.58
MotorcycleE5.804.993.156.393.418.317.72
Piano5.545.699.346.687.178.095.20
PianoL8.9717.5022.9028.4021.1032.4034.40
Pipes7.445.476.7810.607.239.677.53
Playroom8.7612.9012.509.089.3614.005.05
Playtable22.4014.809.705.0929.4024.5013.00
PlaytableP3.473.267.645.187.945.325.67
Recycle6.934.996.273.863.805.563.37
Shelves8.2616.4022.309.7314.7016.209.49
Teddy2.292.641.523.643.514.152.15
Vintage13.1010.4052.6010.7039.7015.009.64
Australia13.406.535.3213.5011.0012.308.48
AustraliaP5.273.365.4812.706.756.625.70
Bicycle24.885.047.7011.007.0111.2010.50
Classroom219.3019.305.6017.5013.7016.305.35
Classroom2E66.5045.7012.5041.8021.5062.6064.80
Computer6.063.418.0511.005.906.833.92
Crusade15.6051.3015.1070.106.7234.0019.20
CrusadeP13.7046.4013.1072.305.8530.6015.30
Djembe1.941.521.844.012.783.652.23
DjembeL13.3029.2016.1028.3022.2037.0019.20
Hoops18.2039.2022.8025.8017.2035.0016.90
Livingroom9.628.7719.407.8811.9013.406.19
Newkuba12.808.1912.6021.5011.1014.207.60
Plants35.4016.9014.5019.7014.1019.1017.90
Staircase19.1027.6027.5021.7023.8034.4024.40
Weighted average12.4315.1011.0017.3610.2315.8510.36

查看原文

表 6全部区域的RMSE对比

Table6. Comparison of RMSE in all regionspixel

Image nameLE-ELASIEBIMstSPSSM-AWPDSGCADoGGuidedProposed algorithm
Adirondack25.4058.9023.0030.3025.1048.9019.20
ArtL16.8038.8039.9038.7048.6059.5022.80
Jadeplant120.00128.0095.80119.00102.00118.0062.60
Motorcycle23.3020.1029.1033.8032.5039.4023.20
MotorcycleE24.0027.4029.8036.9032.4044.6028.90
Piano13.9013.9032.1019.3029.2031.6012.40
PianoL18.7039.0053.8056.3050.8065.8065.30
Pipes30.7030.2041.0047.3047.4052.6035.10
Playroom25.9038.0059.5047.8058.3066.7027.60
Playtable52.3038.6043.3025.0068.3059.5040.10
PlaytableP12.6014.2037.2025.5038.4034.4028.00
Recycle17.6017.5026.5022.5023.3025.7010.60
Shelves15.3030.4044.4023.8033.5033.9019.40
Teddy8.3218.6027.8025.9034.9036.7013.60
Vintage27.4028.80131.0045.80105.0066.1027.70
Australia34.9027.9030.0046.6040.8045.2032.60
AustraliaP26.6022.9030.9045.8035.9038.3027.50
Bicycle221.0023.5032.1037.7032.6040.2032.30
Classroom255.8068.7060.9064.1082.6085.4027.30
Classroom2E112.00106.0076.4093.1088.30148.00127.00
Computer28.2034.1053.7042.3046.7048.0031.70
Crusade58.70156.00141.00152.00131.00151.0079.40
CrusadeP59.00160.00140.00156.00134.00150.0080.10
Djembe8.107.3015.0027.8019.7021.509.28
DjembeL31.0058.6042.8061.3051.4066.4042.70
Hoops51.7077.9078.2072.6073.7099.9050.10
Livingroom23.0024.9067.9039.8052.8058.6017.60
Newkuba53.3038.1065.2078.2068.0081.9033.70
Plants72.6054.6059.3063.9062.9069.4055.70
Staircase46.0048.2085.8073.6078.70102.0051.20
Weighted average36.2548.1554.8555.2056.1564.5035.70

查看原文

表 7非遮挡区域的RMSE对比

Table7. Comparison of RMSE in non-occluded regionspixel

Image nameLE-ELASIEBIMstSPSSM-AWPDSGCADoGGuidedProposed algorithm
Adirondack24.6058.0016.4021.3013.2042.2015.80
ArtL13.8015.6017.2022.7017.9028.3014.10
Jadeplant91.90121.0075.7081.5063.6075.7046.80
Motorcycle18.3011.4015.5019.1014.9021.4014.80
MotorcycleE20.0020.9015.5024.0014.5028.8022.30
Piano13.1012.7027.5014.5020.2019.4010.50
PianoL17.9037.4052.0056.2047.3060.9066.80
Pipes23.3019.2024.5032.2024.6029.6024.50
Playroom25.5031.5035.8027.2025.8037.5013.90
Playtable51.2032.4033.1014.3063.9052.2033.70
PlaytableP11.309.2029.1015.3025.6015.5015.30
Recycle17.1016.9021.9012.9013.6018.108.36
Shelves15.1029.4043.5020.2029.8029.7018.60
Teddy6.8610.706.6312.1014.4014.209.72
Vintage26.3029.00134.0026.20104.0050.7025.00
Australia31.5021.3022.4037.8031.0035.4025.20
AustraliaP22.8016.4023.5036.7024.9025.9022.40
Bicycle217.1017.4026.0029.9022.0030.7027.30
Classroom249.8057.5026.1048.5047.1052.6019.50
Classroom2E112.0099.2053.3086.3059.10136.00126.00
Computer16.609.4730.0022.5017.5020.5011.30
Crusade49.80131.0076.70127.0032.9084.2051.00
CrusadeP48.60123.0071.20131.0031.1079.2037.90
Djembe5.735.549.1317.9011.8014.007.17
DjembeL30.8059.2041.7055.0049.2065.1043.00
Hoops47.5072.7059.7060.1046.3073.2043.40
Livingroom22.7024.3051.8021.3032.6035.3016.00
Newkuba52.0026.5054.0073.8042.5050.5027.90
Plants62.4045.0042.4046.6042.9047.9044.20
Staircase42.0041.8075.4056.2060.6066.2043.50
Weighted average31.4539.1537.4541.2031.4541.5026.65

查看原文

孔令寅, 朱江平, 应三丛. 基于引导图像和自适应支持域的立体匹配[J]. 光学学报, 2020, 40(9): 0915001. Lingyin Kong, Jiangping Zhu, Sancong Ying. Stereo Matching Based on Guidance Image and Adaptive Support Region[J]. Acta Optica Sinica, 2020, 40(9): 0915001.

本文已被 10 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!