Adaptive Weighted AD-Census Noise-Resistant Stereo Matching Algorithm Based on Cross-Support Arm

Sidong Cui; Lei Zhang; Haifeng Zhang; Fengying Yue; Zhichao Yue; Yue Wang; Wenhao Li

doi:10.3788/LOP242334

Laser & Optoelectronics Progress, Volume. 62, Issue 12, 1215013(2025)

Adaptive Weighted AD-Census Noise-Resistant Stereo Matching Algorithm Based on Cross-Support Arm

Sidong Cui^1,2, Lei Zhang³, Haifeng Zhang^2、*, Fengying Yue¹, Zhichao Yue², Yue Wang^1,2, and Wenhao Li^1,2

Author Affiliations

¹School of Electrical and Control Engineering, North University of China, Taiyuan 030051, Shanxi , China

²Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, Xi'an 710119, Shaanxi , China

³Beijing Institute of Astronautical Systems Engineering, Beijing 100076, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Diagram of improved stereo matching algorithm

Download full size

Fig. 2. Rectangular and circular matching templates

Download full size

Fig. 3. Comparison of Census transform disparity maps before and after improvement. (a) Original images; (b) images with noise; (c) Census transform images; (d) improved Census transform images

Download full size

Fig. 4. Schematic of cross-arm extension

Download full size

Fig. 5. Line charts of average mismatch rate. (a)(c) Non-occlusion region; (b)(d) discontinuous disparity region

Download full size

Fig. 6. Middlebury standard image experiment results. (a) Original images; (b) real disparity images; (c) disparity images of proposed algorithm; (d) depth images of proposed algorithm

Download full size

Fig. 7. Boundary profile processing results

Download full size

Fig. 8. Mismatch rates of different algorithms. (a) Non-occlusion region; (b) discontinuous disparity region; (c) mismatch rate of all pixels; (d) average mismatch rate

Download full size

Fig. 9. Test diagrams of simulated planetary surface

Download full size

Table 1. Comparison of various indicators before and after improvement
View table
Table 1. Comparison of various indicators before and after improvement
Algorithm Test image Nocc /% Disc /% PSNR
Census Bowling 20.3 ± 1.5 29.2 ± 2.2 10.629
Wood 15.4 ± 1.1 22.1 ± 1.8 12.817
ImprovedCensus Bowling 10.8 ± 0.4 17.7 ± 0.7 15.825
Wood 9.6 ± 0.3 12.4 ± 0.5 19.006

Table 2. Parameter settings
View table
Table 2. Parameter settings
λ_Census λ_Canny λ_Cs λ_AD τ₁ τ₂ L₁ L₂ $γ_{L}$
15 7 12 21 21 7 30 15 1

Table 3. Average mismatch rates of non-occlusion regions under different noise levels by different algorithms

View table

Table 3. Average mismatch rates of non-occlusion regions under different noise levels by different algorithms

Algorithm	Noiseless	Average mismatch rate /%
		Gaussian noise variance			Salt and pepper noise density
		5	10	15	5%	10%	15%
SGM	9.4±0.3	21.8±1.2	30.4±2.1	43.6±2.8	10.5±0.4	18.2±1.2	28.6±1.8
SNCC	6.5±0.2	20.4±1.0	28.6±1.8	40.9±2.5	8.4±0.2	17.6±1.0	24.4±1.2
AD-Census	7.5±0.4	19.4±1.1	35.6±2.3	49.4±3.2	9.5±0.3	20.5±1.5	35.3±2.2
ELAS	8.8±0.3	27.5±1.5	36.4±2.0	45.3±2.0	10.1±0.4	21.4±1.8	43.7±3.2
ACVNet	4.6±0.3	14.3±1.4	26.3±1.7	38.9±2.1	7.4±0.5	15.6±1.1	25.9±1.9
BGNet	5.4±0.2	16.2±1.2	27.6±2.1	39.5±2.3	8.9±0.4	16.8±1.5	26.7±1.4
Ours	5.6±0.2	15.4±0.8	25.9±1.4	34.8±2.1	7.6±0.2	12.9±0.9	20.2±1.5

Table 4. Average mismatch rates of discontinuous disparity regions under different noise levels by different algorithms

View table

Table 4. Average mismatch rates of discontinuous disparity regions under different noise levels by different algorithms

Algorithm	Noiseless	Average mismatch rate /%
		Gaussian noise variance			Salt and pepper noise density
		5	10	15	5%	10%	15%
SGM	12.5±0.4	17.9±1.0	31.2±2.0	42.7±2.3	15.6±1.2	24.6±1.5	33.2±2.0
SNCC	10.7±0.3	18.5±1.1	30.3±1.9	39.5±2.0	11.5±1.0	19.6±1.2	28.4±1.8
AD-Census	9.5±0.3	25.5±1.4	39.5±2.3	50.9±3.1	13.5±1.3	23.5±1.8	40.3±2.7
ELAS	10.9±0.4	26.6±1.5	37.4±2.2	49.6±2.7	12.9±0.9	23.4±1.1	45.4±2.4
ACVNet	6.5±0.2	15.6±1.2	30.9±1.6	40.5±2.4	10.6±1.1	20.5±1.7	34.6±2.1
BGNet	7.9±0.4	16.8±1.4	32.1±2.1	41.6±2.6	12.7±1.2	21.6±1.5	32.7±1.9
Ours	8.4±0.3	12.5±0.7	27.6±1.5	35.3±2.2	8.9±0.7	15.3±0.9	22.2±1.4

Table 5. Mismatch rates of different algorithms

View table

Table 5. Mismatch rates of different algorithms

Algorithm	Cone			Cloth			Reindeer			Wood			Avg
Algorithm	Nocc	Disc	All	Nocc	Disc	All	Nocc	Disc	All	Nocc	Disc	All	Avg
BSM^［22］	18.4	20.8	19.6	17.5	19.5	18.6	23.5	23.9	23.7	15.9	16.7	15.7	19.2
PPEP-GF^［23］	11.8	14.5	12.7	13.4	15.8	15.3	17.2	21.3	18.6	13.6	15.9	13.8	15.3
Z2ZNCC^［24］	15.6	16.6	15.5	14.7	17.6	14.9	18.8	21.7	19.1	15.5	14.8	16.2	16.8
SGBM1^［17］	14.9	15.8	13.8	13.7	14.6	14.2	16.7	16.4	17.4	12.6	12.7	11.3	14.6
AD-Census^［6］	11.3	14.4	13.4	12.4	15.4	12.8	14.7	18.1	15.7	12.4	16.4	14.5	14.1
SGM^［17］	10.5	12.5	11.4	11.9	11.6	10.3	12.6	14.7	15.9	12.5	13.8	11.9	12.1
r200high^［25］	21.2	22.3	22.1	20.6	19.8	22.4	23.2	26.2	24.4	23.3	24.8	23.7	21.4
MC-CNN^［26］	8.4	9.5	8.9	7.6	9.4	8.7	12.9	13.4	11.1	7.3	7.8	6.8	9.1
SNCC^［18］	9.7	13.8	12.8	10.5	11.3	9.4	13.1	15.2	17.2	8.9	11.7	9.8	11.9
IGEV++^［27］	2.4	4.0	3.8	2.7	3.6	3.3	2.9	3.9	3.1	2.8	3.4	2.5	3.2
MoCha-V2^［28］	2.8	4.6	3.1	3.5	4.7	4.0	2.3	5.2	3.9	3.7	4.4	3.8	3.5
Ours	7.6	9.1	8.3	5.5	7.6	6.3	8.7	11.7	10.4	6.4	7.5	7.3	8.7

Table 6. Running time of each algorithm
View table
Table 6. Running time of each algorithm
Algorithm BSM PPEP-GF Z2ZNCC SGBM1 AD-Census MC-CNN SNCC Ours
Time /s 3.14 2.44 2.57 3.64 2.34 1.68 2.42 1.42

Table 7. Comparison of measured and actual distances

View table

Table 7. Comparison of measured and actual distances

No.	Centroid coordinate	Measured distance /mm	Actual distance /mm	Error /%
1	（136， 242）	677.5	689.4	1.70
2	（307， 148）	765.3	749.1	2.10
3	（431， 426）	482.4	486.2	0.87
4	（564， 156）	760.5	772.7	1.60
5	（620， 78）	834.3	813.5	2.50

Tools

Get Citation

Copy Citation Text

Sidong Cui, Lei Zhang, Haifeng Zhang, Fengying Yue, Zhichao Yue, Yue Wang, Wenhao Li. Adaptive Weighted AD-Census Noise-Resistant Stereo Matching Algorithm Based on Cross-Support Arm[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1215013

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites