Laser & Optoelectronics Progress, Volume. 59, Issue 16, 1615006(2022)
Matching Algorithm Based on Improved Cost Calculation and Path Optimization Strategy
Figures & Tables(12)
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
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
Table 1. Parameters of proposed algorithm
View tableTable 1. Parameters of proposed algorithm
Parameter LBP size P1 P2 R ξ ζ λ Value 7×7 10 150 5 10 2 5
Table 2. Calculation time comparison in debug mode
View tableTable 2. Calculation time comparison in debug mode
Algorithm Tsukuba Venus Teddy Cones d∈［0，16］ d∈［0，20］ d∈［0，64］ d∈［0，64］ Cost calculation Cost aggregation Cost calculation Cost aggregation Cost calculation Cost aggregation Cost calculation Cost aggregation SGM 0.170 8.472 0.280 15.777 0.695 49.987 0.713 50.114 Proposed 0.053 5.354 0.074 10.076 0.253 29.979 0.254 29.714 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 3. Calculation time comparison in release mode
View tableTable 3. Calculation time comparison in release mode
Algorithm Tsukuba Venus Teddy Cones d∈［0，16］ d∈［0，20］ d∈［0，64］ d∈［0，64］ Cost calculation Cost aggregation Cost calculation Cost aggregation Cost calculation Cost aggregation Cost calculation Cost aggregation SGM 0.037 0.088 0.071 0.163 0.196 0.445 0.205 0.493 Proposed 0.017 0.045 0.025 0.097 0.075 0.311 0.093 0.289 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 4. Comparison of mismatched ratio with threshold is 1
View tableTable 4. Comparison of mismatched ratio with threshold is 1
Algorithm Tsukuba Venus Teddy Cones Average nonocc all disc nonocc all disc nonocc all disc nonocc all disc RTcensus［ 19 ］5.1 6.3 19.2 1.6 2.4 14.2 8.0 13.8 20.3 4.1 9.5 12.2 9.73 ADcensus［ 20 ］1.1 1.5 5.7 0.1 0.3 1.2 4.1 6.2 10.9 2.4 7.3 7.0 3.97 C-SemiGlob［ 23 ］2.6 3.3 9.9 0.3 0.6 3.2 5.1 11.8 13.0 2.8 8.4 8.2 5.76 planeFitSGM［ 21 ］3.1 4.2 14.9 1.1 1.9 14.6 5.7 11.6 17.1 3.8 9.3 11.3 8.21 SGM［ 14 ］3.3 4.0 12.8 1.0 1.6 11.3 6.0 12.2 16.3 3.1 9.8 8.9 7.51 SGMDDW［ 22 ］2.3 4.4 11.8 1.2 2.7 16.8 6.5 14.5 17.5 5.6 14.2 14.8 9.36 Proposed 3.3 3.5 12.7 0.9 1.1 9.8 7.3 9.1 14.4 3.9 7.3 11.7 7.08
Table 5. Comparison of mismatched ratio when threshold is 0.5
View tableTable 5. Comparison of mismatched ratio when threshold is 0.5
Algorithm Tsukuba Venus Teddy Cones Average nonocc all disc nonocc all disc nonocc all disc nonocc all disc RTcensus［ 19 ］12.9 14.1 28.1 3.7 4.6 17.8 11.4 18.6 27.7 5.5 11.8 15.9 14.34 ADcensus［ 20 ］26.8 27.0 21.1 4.1 4.6 8.0 10.6 13.8 20.1 6.6 12.4 11.9 13.92 C-SemiGlob［ 23 ］13.9 14.7 18.9 3.3 3.8 10.9 9.8 17.4 22.8 5.4 11.7 12.8 12.12 planeFitSGM［ 21 ］9.2 10.4 23.3 2.3 3.2 15.9 9.0 17.0 25.5 5.2 11.6 14.7 12.28 SGM［ 14 ］13.4 14.3 20.3 4.6 5.4 15.7 11.0 18.5 26.1 4.9 12.5 13.5 13.35 SGMDDW［ 22 ］15.5 17.5 23.1 8.1 9.6 25.6 15.7 23.9 31.5 14.5 22.8 26.0 19.48 Proposed 6.3 8.9 20.1 5.5 5.9 17.3 8.2 10.0 30.5 8.5 10.1 17.4 12.39
Table 6. Comparison of mismatched ratio in all regions
View tableTable 6. Comparison of mismatched ratio in all regions
Stereo
pairs
Mismatched ratio MDP［ 23 ］SRM［ 24 ］Cens5［ 25 ］SGBM1［ 13 ］SGM［ 14 ］DTS［ 26 ］Proposed Adiron 25.0 28.7 37.0 39.5 29.1 20.9 25.5 ArtL 15.1 15.8 17.5 19.0 11.5 17.7 11.7 Jadepl 35.9 38.2 35.9 36.0 28.1 34.3 20.2 Motor 30.4 27.6 32.5 25.9 25.5 21.1 17.9 MotorE 28.7 30.3 28.9 36.9 22.5 21.8 15.8 Piano 33.6 38.1 33.6 36.6 26.1 33.0 22.3 PianoL 41.7 47.9 47.4 58.6 42.1 39.0 39.5 Pipes 29.0 23.2 23.6 20.4 20.5 24.9 14.4 Playrm 42.7 44.6 47.4 45.0 38.3 40.6 26.8 Playt 47.8 35.0 67.9 52.4 64.7 34.7 45.3 PlaytP 28.8 32.9 29.1 29.2 24.7 32.2 17.3 Recyc 31.3 32.4 33.6 33.0 27.0 24.1 18.9 Shelvs 54.5 54.4 62.1 57.2 59.0 46.9 41.3 Teddy 10.8 10.6 12.3 18.1 10.1 10.3 8.2 Vintge 43.6 49.5 61.2 55.0 51.3 47.6 35.9 Austr 36.3 29.3 63.2 51.8 58.1 24.5 40.7 AustrP 18.4 27.2 19.0 24.1 10.2 23.6 17.1 Bicyc2 24.4 25.6 19.2 21.4 15.5 17.5 17.9 Class 30.0 29.5 33.7 24.4 30.4 21.7 31.3 ClassE 48.8 40.8 57.3 78.5 54.6 38.3 38.2 Compu 16.9 18.9 30.0 20.8 21.2 20.9 17.9 Crusa 40.7 45.5 61.3 47.8 48.8 44.3 34.2 CrusaP 32.8 45.2 47.7 41.9 27.9 45.0 19.5 Djemb 20.9 21.8 19.2 13.4 13.9 19.2 23.7 DjembL 47.3 42.5 49.7 63.5 44.1 33.8 30.9 Hoops 41.2 41.5 51.2 49.8 44.2 39.3 31.9 Livgrm 28.8 33.4 39.5 34.6 33.1 29.1 23.2 Nkuba 38.3 38.7 36.4 34.6 32.4 36.1 20.7 Plants 33.5 32.3 43.0 35.4 31.9 31.7 22.3 Stairs 35.2 42.5 55.0 59.2 44.7 29.8 31.3 Average 30.9 32.4 36.6 35.2 29.7 28.6 25.4
Tools
Get Citation
Copy Citation Text
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
Paper Information
Category: Machine Vision
Received: May. 18, 2021
Accepted: Aug. 3, 2021
Published Online: Jul. 22, 2022
The Author Email: Kangsheng Lai (laiksh@dlut.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20