基于多特征SAD-Census变换的立体匹配算法

吴福培; 黄耿楠; 刘宇豪; 叶玮琳; 李昇平

doi:10.37188/CO.2023-0082

基于多特征SAD-Census变换的立体匹配算法

doi: 10.37188/CO.2023-0082

汕头大学工学院机械工程系, 广东汕头 515063

基金项目: 国家自然科学基金（No. 61573233）；广东省自然科学基金（No. 2021A1515010661）；广东省普通高校重点领域专项（No. 2020ZDZX2005）

详细信息

作者简介:
吴福培（1980—），男，广西玉林人，博士，教授，2009年于华南理工大学机械工程专业获得博士学位，现就职于汕头大学机械工程系，主要研究方向：自动光学检测和3D测量。E-mail：fpwu@stu.edu.cn

黄耿楠（1997—），男，广东揭阳人，硕士， 2023年于汕头大学机械工程专业获得硕士学位，现就职于比亚迪有限公司，担任高级工程师。研究方向：双目视觉与三维测量。E-mail：907877156@qq.com

刘宇豪（1999—），男，重庆人，硕士研究生，主要研究方向：双目视觉与三维测量。E-mail：21yhliu@stu.edu.cn

叶玮琳（1984—），女，广东潮州人，副教授，博士，主要研究方向：光电传感技术。E-mail：wlye@stu.edu.cn

李昇平（1966—），男，湖南永州人，教授，博士，主要研究方向：自适应控制、机器视觉。E-mail：spli@stu.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 348
- HTML全文浏览量: 126
- PDF下载量: 171
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-08
- 修回日期: 2023-06-02
- 网络出版日期: 2023-09-15

Stereo matching algorithm based on multi-feature SAD-Census transformation

Department of Mechanical Engineering, College of Engineering, Shantou University, Shantou 515063, China

Funds: Supported by National Natural Science Foundation of China (No. 6l573233); Natural Science Foundation of Guangdong Province (No. 2021A1515010661); Special Projects in Key Areas of General Universities in Guangdong Province (No. 2020ZDZX2005)

More Information

Corresponding author: 907877156@qq.com

摘要

摘要:
视差不连续区域和重复纹理区域的误匹配率高一直是影响双目立体匹配测量精度的主要问题，为此，本文提出一种基于多特征融合的立体匹配算法。首先，在代价计算阶段，通过高斯加权法赋予邻域像素点的权值，从而优化绝对差之和（Sum of Absolute Differences，SAD）算法的计算精度。接着，基于Census变换改进二进制链码方式，将邻域内像素的平均灰度值与梯度图像的灰度均值相融合，进而建立左右图像对应点的判断依据并优化其编码长度。然后，构建基于十字交叉法与改进的引导滤波器相融合的聚合方法，从而实现视差值再分配，以降低误匹配率。最后，通过赢家通吃（Winner Take All，WTA）算法获取初始视差，并采用左右一致性检测方法及亚像素法提高匹配精度，从而获取最终的视差结果。实验结果表明，在Middlebury数据集的测试中，所提SAD-Census算法的平均非遮挡区域和全部区域的误匹配率为分别为2.67%和5.69%，测量200~900 mm距离的平均误差小于2%；而实际三维测量的最大误差为1.5%。实验结果检验了所提算法的有效性和可靠性。
- 机器视觉 /
- 立体匹配 /
- SAD-Census变换 /
- 十字交叉法 /
- 引导滤波
Abstract:
The high mismatching rate of the parallax discontinuity region and the repeated texture region has been a major issue affecting the measurement accuracy of binocular stereo matching. For these reasons, we propose a stereo matching algorithm based on multi-feature fusion. Firstly, the weight of neighboring pixels is given using Gaussian weighting method, which optimizes the calculation accuracy of the Sum of Absolute Differences (SAD) algorithm. Based on the Census transformation, the binary chain code technique has been enhanced to fuse the average gray value of neighborhood pixels with the average gray value of gradient image, and then the judgment basis of the left and right image corresponding points is established, and the coding length is optimized. Secondly, an aggregation technique has been developed that combines the cross method and the improved guide filter to redistribute disparity values with the aim of minimizing false matching rate. Finally, the initial disparity is obtained by the Winner Take All (WTA) algorithm, and the final disparity results are obtained by the left-right consistency detection method, sub-pixel method, and then a stereo matching algorithm based on the multi-feature SAD Census transform is established. The experimental results show that in the testing of the Middlebury dataset, the average mismatch rates of the proposed algorithm for non-occluded regions and all regions are 2.67% and 5.69%, the average error of the 200−900 mm distance is less than 2%, and the maximum error of the actual 3D data measurement is 1.5%. Experimental results verify the effectiveness of the proposed algorithm.
- machine vision /
- stereo matching /
- SAD-Census transform /
- cross method /
- guided filtering

HTML全文

图 1 改进Census变换示意图

Figure 1. Schematic diagram of improved Census transform

下载: 全尺寸图片幻灯片

图 2 自适应十字交叉域

Figure 2. Adaptive crossing domain

下载: 全尺寸图片幻灯片

图 3 优化聚合路径示意图

Figure 3. Schematic diagram of optimized aggregation path

下载: 全尺寸图片幻灯片

图 4 3种代价计算方法的视差图。（a）传统SAD-Census算法；（b）传统SAD+改进Census算法；（c）本文算法

Figure 4. Parallax maps of three cost calculation methods. (a) Traditional SAD-Census algorithms; (b) traditional SAD + improved Census algorithm; (a) the proposed algorithm

下载: 全尺寸图片幻灯片

图 5 4种算法的实验结果图

Figure 5. Experimental results of 4 kinds of algorithms

下载: 全尺寸图片幻灯片

图 6 不同算法的误匹配率

Figure 6. Mismatch rate of different algorithms

下载: 全尺寸图片幻灯片

图 7 不同亮度下几种算法的实验结果

Figure 7. Experimental results of 4 kinds of algorithms at different brightness levels

下载: 全尺寸图片幻灯片

图 8 噪声干扰下各算法的实验结果

Figure 8. Experimental results of different algorithms under noise interference

下载: 全尺寸图片幻灯片

图 9 测距实验对比结果

Figure 9. Comparison results of distance measuring experiments

下载: 全尺寸图片幻灯片

图 10 样本视差结果图的边缘提取

Figure 10. Edge extraction of sample parallax result maps

下载: 全尺寸图片幻灯片

图 11 量块视差图

Figure 11. Gauge block parallax map

下载: 全尺寸图片幻灯片

表 1 实验参数设置

Table 1. Experimental parameter setting

Parameter	Value	Parameter	Value
б	2	λ	0.003
τ₁	16	ε	0.001
τ₂	4	ω	9×7
L₁	30	β₁	0.6
L₂	15	β₂	0.4
δ₀	1

下载: 导出CSV

表 2 不同算法在非遮挡区域和全部区域的误匹配率

Table 2. Mismatch rate of different algorithms in the non-occluded region and all regions (%)

Algorithm	Tsukuba			Venus			Teddy			Cones			Avg
Algorithm	Non	all	Disc	Non	all	Disc	Non	all	Disc	Non	all	Disc	Non	all	Disc
GC-occ	1.19	2.01	6.24	1.64	2.19	6.75	11.20	17.40	19.80	5.36	12.40	13.00	4.84	8.50	11.45
SemiGlob	3.26	3.96	12.80	1.00	1.57	11.30	6.02	12.20	16.30	3.06	9.75	8.90	3.34	6.87	12.32
RTCensus	5.08	6.25	19.20	1.58	2.42	14.20	7.96	13.8	20.30	4.10	9.54	12.20	4.68	8.01	16.48
HCA	1.31	2.47	0.94	0.11	0.23	0.24	4.23	9.45	3.67	5.64	11.59	4.30	2.83	5.94	2.29
Cross-ScaleGF	2.38	2.85	8.40	1.13	1.98	9.26	7.05	14.9	16.80	3.32	11.00	7.99	3.47	7.68	10.61
GradAdaptWgt	2.26	2.63	8.99	0.99	1.39	4.92	8.00	13.10	18.60	2.61	7.67	7.43	3.46	6.19	9.99
Proposed algorithm	1.25	2.73	0.90	0.15	0.31	0.31	4.62	8.73	2.98	4.69	10.98	4.11	2.67	5.69	2.08

下载: 导出CSV

表 3 三维测距实验结果

Table 3. Experimental results of three-dimensional distance measurement

序号	实际距离/mm	测量距离/mm	距离误差/mm	距离误差率/%
1	200	198.42	−1.58	0.79
2	300	302.55	2.55	0.85
3	400	403.76	3.76	0.94
4	500	495.05	−4.95	0.99
5	600	606.06	6.06	1.01
6	700	707.56	7.56	1.08
7	800	809.52	9.52	1.19
8	900	912.24	12.24	1.36

下载: 导出CSV

表 4 测量样本实验结果

Table 4. Sample experimental results

样本序号	实际长度/mm	测量长度/mm	误差/mm	误差率/%
1	20	19.9575	−0.0425	−0.21
1	35	34.6144	−0.3856	−1.1
2	35	35.1649	0.1649	−0.47
2	50	49.9672	−0.0328	−0.06
3	5	4.9439	−0.0561	−1.12
3	15	14.7565	−0.2435	−1.62
4	20	20.3541	0.3541	0.71
4	30	29.5784	−0.4216	−1.41

下载: 导出CSV

表 5 厚度测量结果

Table 5. Thickness measurement results

序号	实际厚度/mm	测量厚度/mm	测量误差/mm	绝对误差率/%
1	1.00	1.0105	0.0105	1.05
2	1.10	1.1235	0.0235	2.13
3	1.20	1.2335	0.0335	2.79
4	1.50	1.5563	0.0563	3.75
5	2.00	1.9346	−0.0654	3.27

下载: 导出CSV

参考文献(30)

[1]	陈仁虹, 梁晋, 叶美图, 等. 柔性复合薄膜成形极限曲线的视觉测定方法[J]. 中国光学,2022,15(1):22-33. doi: 10.37188/CO.2021-0101 CHEN R H, LIANG J, YE M T, et al. Visual method for measuring forming limit curve of pliable composite film[J]. Chinese Optics, 2022, 15(1): 22-33. (in Chinese) doi: 10.37188/CO.2021-0101
[2]	任晓奎, 关钧渤, 殷新勇, 等. 基于改进局部一致性约束的立体匹配算法[J]. 液晶与显示,2023,38(4):543-553. doi: 10.37188/CJLCD.2022-0226 REN X K, GUAN J B, YIN X Y, et al. Stereo matching algorithm based on improved local consistency constraint[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(4): 543-553. (in Chinese) doi: 10.37188/CJLCD.2022-0226
[3]	张强, 刘婷婷, 李海滨, 等. 基于最佳搜索域的水下图像区域匹配算法研究[J]. 光学学报,2014,34(6):0615001. doi: 10.3788/AOS201434.0615001 ZHANG Q, LIU T T, LI H B, et al. Research on region matching for underwater images based on optimum searching area[J]. Acta Optica Sinica, 2014, 34(6): 0615001. (in Chinese) doi: 10.3788/AOS201434.0615001
[4]	LEE J, JUN D, EEM C, et al. Improved census transform for noise robust stereo matching[J]. Optical Engineering, 2016, 55(6): 063107. doi: 10.1117/1.OE.55.6.063107
[5]	彭新俊, 韩军, 汤踊, 等. 基于改进Census变换和异常值剔除的抗噪立体匹配算法[J]. 光学学报,2017,37(11):1115004. doi: 10.3788/AOS201737.1115004 PENG X J, HAN J, TANG Y, et al. Anti-noise stereo matching algorithm based on improved census transform and outlier elimination[J]. Acta Optica Sinica, 2017, 37(11): 1115004. (in Chinese) doi: 10.3788/AOS201737.1115004
[6]	刘建国, 俞力, 柳思健, 等. 基于改进Census变换和多尺度空间的立体匹配算法[J]. 华南理工大学学报(自然科学版),2017,45(12):43-49. LIU J G, YU L, LIU S J, et al. Stereo matching algorithm based on improved census transform and multi-scale space[J]. Journal of South China University of Technology (Natural Science Edition), 2017, 45(12): 43-49. (in Chinese)
[7]	杨春雨, 顾振, 张鑫, 等. 基于深度学习的带式输送机煤流量双目视觉测量[J]. 仪器仪表学报,2021,42(8):164-174. YANG CH Y, GU ZH, ZHANG X, et al. Binocular vision measurement of coal flow of belt conveyors based on deep learning[J]. Chinese Journal of Scientific Instrument, 2021, 42(8): 164-174. (in Chinese)
[8]	HOSNI A, BLEYER M, GELAUTZ M. Secrets of adaptive support weight techniques for local stereo matching[J]. Computer Vision and Image Understanding, 2013, 117(6): 620-632. doi: 10.1016/j.cviu.2013.01.007
[9]	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. doi: 10.3788/AOS202040.0915001 KONG L Y, ZHU J P, YING S C. Stereo matching based on guidance image and adaptive support region[J]. Acta Optica Sinica, 2020, 40(9): 0915001. (in Chinese) doi: 10.3788/AOS202040.0915001
[10]	CHANG N Y C, TSAI T H, HSU B H, et al. Algorithm and architecture of disparity estimation with mini-census adaptive support weight[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2010, 20(6): 792-805. doi: 10.1109/TCSVT.2010.2045814
[11]	王森, 危辉, 孟令江. 基于控制点和RGB向量差联合梯度Census变换的立体匹配算法[J]. 模式识别与人工智能,2022,35(1):37-50. WANG S, WEI H, MENG L J. Stereo matching algorithm based on control points, RGB vector difference and gradient census transform[J]. Pattern Recognition and Artificial Intelligence, 2022, 35(1): 37-50. (in Chinese)
[12]	ZHANG K, LU J B, LAFRUIT G. Cross-based local stereo matching using orthogonal integral images[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2009, 19(7): 1073-1079. doi: 10.1109/TCSVT.2009.2020478
[13]	WANG W Q, YAN J, XU N Y, et al. Real-time high-quality stereo vision system in FPGA[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2015, 25(10): 1696-1708. doi: 10.1109/TCSVT.2015.2397196
[14]	CHENG F Y, ZHANG H, SUN M G, et al. Cross-trees, edge and superpixel priors-based cost aggregation for stereo matching[J]. Pattern Recognition, 2015, 48(7): 2269-2278. doi: 10.1016/j.patcog.2015.01.002
[15]	洪梓嘉, 李彦明, 林洪振, 等. 基于双目视觉的种植前期农田边界距离检测方法[J]. 农业机械学报,2022,53(5):27-33,56. HONG Z J, LI Y M, LIN H ZH, et al. Field boundary distance detection method in early stage of planting based on binocular vision[J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53(5): 27-33,56. (in Chinese)
[16]	ZHAN Y L, GU Y Z, ZHANG X L, et al. Stereo matching based on efficient image-guided cost aggregation[J]. IEICE Transactions on Information and Systems, 2016, E99.D(3): 781-784. doi: 10.1587/transinf.2015EDL8223
[17]	YAO M, OUYANG W B, XU B G. Hybrid cost aggregation for dense stereo matching[J]. Multimedia Tools and Applications, 2020, 79(31): 23189-23202.
[18]	YANG Q X. Stereo matching using tree filtering[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(4): 834-846. doi: 10.1109/TPAMI.2014.2353642
[19]	HIRSCHMÜLLER H. Accurate and efficient stereo processing by semi-global matching and mutual information[C]. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, IEEE, 2005: 807-814.
[20]	张建业, 朴燕. 基于改进稳态匹配概率的立体匹配算法研究[J]. 液晶与显示,2018,33(4):357-364. doi: 10.3788/YJYXS20183304.0357 ZHANG J Y, PIAO Y. Stereo matching algorithm based on improved steady-state matching probability[J]. Chinese Journal of Liquid Crystals and Displays, 2018, 33(4): 357-364. (in Chinese) doi: 10.3788/YJYXS20183304.0357
[21]	LI ZH G, ZHENG J H, ZHU Z J, et al. Weighted guided image filtering[J]. IEEE Transactions on Image Processing, 2015, 24(1): 120-129. doi: 10.1109/TIP.2014.2371234
[22]	HUMENBERGER M, ZINNER C, WEBER M, et al. A fast stereo matching algorithm suitable for embedded real-time systems[J]. Computer Vision and Image Understanding, 2010, 114(11): 1180-1202. doi: 10.1016/j.cviu.2010.03.012
[23]	刘杰, 张建勋, 代煜, 等. 基于跨尺度引导图像滤波的稠密立体匹配[J]. 光学学报,2018,38(1):0115004. doi: 10.3788/AOS201838.0115004 LIU J, ZHANG J X, DAI Y, et al. Dense stereo matching based on cross-scale guided image filtering[J]. Acta Optica Sinica, 2018, 38(1): 0115004. (in Chinese) doi: 10.3788/AOS201838.0115004
[24]	KOLMOGOROV V, ZABIH R. Computing visual correspondence with occlusions using graph cuts[C]. IEEE International Conference on Computer Vision, IEEE, 2001: 508-515.
[25]	DE-MAEZTU L, VILLANUEVA A, CABEZA R. Stereo matching using gradient similarity and locally adaptive support-weight[J]. Pattern Recognition Letters, 2011, 32(13): 1643-1651. doi: 10.1016/j.patrec.2011.06.027
[26]	任久斌. 基于双目立体视觉的三维重建技术研究[D]. 成都: 西南交通大学, 2020. REN J B. Research on 3D reconstruction technology based on binocular stereo vision[D]. Chengdu: Southwest Jiaotong University, 2020. (in Chinese)
[27]	符强, 孔健明, 纪元法, 等. 基于双目视觉的无人机实时测距算法[J]. 电光与控制,2023,30(4):94-99. FU Q, KONG J M, JI Y F, et al. A method of UAV real-time ranging based on binocular vision[J]. Electronics Optics & Control, 2023, 30(4): 94-99. (in Chinese)
[28]	仲伟波, 姚旭洋, 冯友兵, 等. 双目区域视差快速计算及测距算法[J]. 中国图象图形学报,2019,24(9):1537-1545. ZHONG W B, YAO X Y, FENG Y B, et al. Rapid calculation and ranging algorithm based on binocular region parallax[J]. Journal of Image and Graphics, 2019, 24(9): 1537-1545. (in Chinese)
[29]	肖纯, 林嘉欣, 王玮. 基于双目测距的晶英石板自动打磨系统研究[J]. 武汉理工大学学报,2018,40(3):78-83. XIAO CH, LIN J X, WANG W. Research on automatic jing ying stone grinding system based on binocular distance measurement[J]. Journal of Wuhan University of Technology, 2018, 40(3): 78-83. (in Chinese)
[30]	HUANG L, WU G P, LIU J Y, et al. Obstacle distance measurement based on binocular vision for high-voltage transmission lines using a cable inspection robot[J]. Science Progress, 2020, 103(3): 003685042093691.