双目视觉技术被广泛应用于生产生活中的多个领域，而立体匹配算法是双目视觉的研究重点与难点，其匹配精度直接影响双目视觉效果，因此对立体匹配算法展开研究具有重要意义。本文主要研究工作如下：

（1）针对局部立体匹配算法存在视差图像边缘模糊及匹配精度较低问题，提出一种改进Census变换与特征融合的立体匹配算法。首先，使用变换窗口的邻域像素信息代替中心像素，解决传统Census变换过度依赖窗口中心像素的问题；其次，引入图像的颜色信息与梯度信息构建融合匹配代价计算函数，以提高初始匹配代价的可靠性；再次，为建立邻域像素点间的联系，引入单向动态规划思想进行匹配代价聚合；最后，提出一种基于八方向的视差空洞填充方法对视差图进行优化。实验结果表明，该算法优于其它改进Census变换算法，具有较高的匹配精度。

（2）针对现有立体匹配算法在匹配过程中忽略了不同纹理区域的影响，导致算法适应性受限问题，提出一种基于纹理分区的立体匹配算法。首先，充分利用图像纹理信息，提出一种基于灰度方差的纹理分区方法将图像划分为强弱纹理区域；其次，根据不同纹理区域的特征构建不同的匹配代价计算策略；再次，对十字交叉域的臂长判定准则以及聚合策略进行改进，以适应不同纹理区域；最后，加入多种视差优化策略对视差图进行优化。实验结果表明，该算法在多种场景下的立体匹配任务中，图像平均误匹配率较低，提高了立体匹配算法的适应性。

Binocular vision technology is widely used in many fields of production and life, and the stereo matching is the focus and difficulty of binocular vision, and its matching accuracy directly affects the binocular visual effect, so the research on stereo matching algorithm is of great significance. The main research work of this paper is as follows:

(1) Aiming at the problems of blurred edges and low matching accuracy of local stereo matching algorithm, a method based on improved Census transform and feature fusion is proposed. Firstly, neighborhood pixel information replaces the central pixel to reduce dependency in traditional Census transform. Secondly, color and gradient information are used to construct a fused cost function, improving matching costs. Thirdly, dynamic programming is introduced for better cost aggregation and pixel connectivity. Finally, an eight-directional disparity hole filling method optimizes the disparity map. Experimental results show that the algorithm outperforms other improved Census transform methods in matching accuracy.

(2) Aiming to address the limited adaptability of stereo matching algorithms due to unaddressed texture variations, a stereo matching algorithm based on texture partitioning is proposed. Firstly, the algorithm uses grayscale variance to partition images into different texture areas. Secondly, different matching cost calculation strategies are constructed according to the characteristics of different texture regions. Thirdly, the arm length determination criterion of the cross domain and the aggregation strategy are improved to adapt to different texture regions. Finally, multiple parallax optimisation strategies are added to optimise the parallax map. Experimental results show that the algorithm achieves a low average image mismatch rate across diverse scenarios, which improves the adaptability of the stereo matching algorithms.

[1] Wang D, Sun H, Lu W, et al. A novel binocular vision system for accurate 3-D reconstruction in large-scale scene based on improved calibration and stereo matching methods[J]. Multimedia Tools and Applications, 2022, 81(18):26265-26281.

[2] 方博文,张晓东,陈敬义等.基于双目视觉的行车中障碍距离检测方法研究[J].机械设计与制造,2019(04):94-98.

[3] Wei H, Meng L. A stereo matching algorithm for high-precision guidance in a weakly textured industrial robot environment dominated by planar facets[J]. Computer Graphics Forum, 2022, 41(1):288-300.

[4] 王正家,景嘉宝,王思宇.基于双目视觉的车辆外廓尺寸测量方法[J].电子测量技术,2023,46(12):150-156.

[5] Marr D. Vision: A computational investigation into the human representation and processing of visual information [J]. The Quarterly Review of Biology, 1982,58(2):1-6.

[6] Zou X, Zou H, Lu J. Virtual manipulator-based binocular stereo vision positioning system and errors modelling[J]. Machine Vision and Applications, 2012, 23(1):43-63.

[7] Barnard S T, Fischler M A. Computational stereo[J]. ACM Computing Surveys(CSUR), 1982, 14(4): 553-572.

[8] Gong M, Yang Y H. Fast unambiguous stereo matching using reliability-based dynamic programming[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2005, 27(6):998-1003.

[9] 曹义,王敬东,李鹏.基于垂直性约束的动态规划立体匹配算法[J].红外技术, 2008,30(12):722-726.

[10] Hu T B, Qi B J, Wu T, et al. Stereo matching using weighted dynamic programming on a single-direction four-connected tree[J]. Computer Vision ＆ Image Understanding, 2012, 116(8) : 908-921.

[11] Hallek M, Boukamcha H, Mtibaa A, et al. Dynamic programming with adaptive and self-adjusting penalty for real-time accurate stereo matching[J]. Journal of Real-Time Image Processing, 2021, 19(2):233-245.

[12] 田茂,花向红.树动态规划的超像素层次立体匹配算法[J].测绘科学, 2021, 46(12):123-128.

[13] Boykov Y, Kolmogorov V. An experimental comparison of min-cut/max-flowalgorithms for energy minimization in vision[J]. IEEE transactions on pattern analysis and machine intelligence, 2004, 26(9): 1124-1137.

[14] Wang D L, Kah B L. Obtaining depth map from segment-based stereo matching using graph cuts[J]. Journal of visual communication & image representation, 2011, 22(4):325-331.

[15] 曹国震, 彭寒. 基于区域一致性的图割立体匹配[J]. 西北工业大学学报, 2017,35(1): 160-163.

[16] Lu B, Sun L, Yu L, et al. An improved graph cut algorithm in stereo matching[J].Displays, 2021, 69:1-7.

[17] Pearl J．Reverend Bayes on inference engines: a distributed hierarchical approach[C]∥Proceedings of the 2nd National Conference on Artificial Intelligence. 1982: 133-136.

[18] 何栿,达飞鹏.置信度传播和区域边缘构建的立体匹配算法[J].中国图象图形学报,2011,16(11):2060-2066.

[19] Wang X, Su Y, Tang L, et al. A combined back and foreground-based stereo matching algorithm using belief propagation and self-adapting dissimilarity measure[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2018, 32(6):1850019.1-1850019.18.

[20] Pan C, Liu Y Q, Huang D P. Novel belief propagation algorithm for stereo matching with a robust cost computation[J]. IEEE Access, 2019, 7: 29699-29708.

[21] Xu Y B, Yu D B, Ma Y P, et al. Underwater stereo-matching algorithm based on belief propagation[J]. Signal Image And Video Processing,2023,17(4): 891-897.

[22] Hirschmuller H. Stereo processing by semiglobal matching and mutual information[J]. IEEE Transactions on pattern analysis and machine intelligence,2007, 30(2): 328-341.

[23] 王阳萍,秦安娜,郝旗等.结合加速鲁棒特征的遥感影像半全局立体匹配[J].光学学报,2020,40(16):163-171.

[24] Toledo J, Lauer M, Stiller C. Real-time stereo semi-global matching for video processing using previous incremental information[J]. Journal of Real-Time Image Processing, 2021, 19(1):205-216.

[25] 李聪聪,方勇,王芮等.顾及图像分割信息的半全局立体匹配算法研究[J].电子测量技术,2022,45(05):140-145.

[26] Deng C G, Liu D Y, Zhang H D, et al. Semi-global stereo matching algorithm based on multi-scale information fusion[J]. Applied Science, 2023, 13(2):1-14.

[27] 刘学君,常梦洁,孔祥旻等.仓库三维重建系统中改进双目匹配SAD算法研究[J].计算机应用与软件,2023,40(07):180-184+214.

[28] Wang X T, Wang X B, Han L L. A Novel Parallel Architecture for Template Matching based on Zero-Mean Normalized Cross-Correlation[J]. IEEE Access, 2019, 7:186626-186636.

[29] Anandan P. A computational framework and an algorithm for the measurement of visual motion[J]. International Journal of Computer Vision, 1989, 2(3): 283-310.

[30] Dinh V Q, Pham C C, Jeon J W. Matching cost function using robust soft rank transformations[J]. Iet Image Processing, 2016, 10(7): 561-569.

[31] Huang C H, Yang J F. Improved quadruple sparse census transform and adaptive multi-shape aggregation algorithms for precise stereo matching[J]. IET Computer Vision, 2021,16(2): 159-179.

[32] Yoon K J, Kweon I S. Adaptive support-weight approach for correspondence search[J]. IEEE transactions on pattern analysis and machine intelligence, 2006,28(4): 650-656.

[33] Li Y J, Zhag J W, Zhong Y Z, et al. An efficient stereo matching based on fragment matching[J]. The Visual Computer, 2018, 35(2):257-269.

[34] 陈先锋,郭正华,伍俊龙等.基于区域先验信息的去遮挡立体匹配算法[J].激光与光电子学进展,2019,56(19):95-101.

[35] Zhang J, Zhang Y, Wang C, et al. Binocular stereo matching algorithm based on MST cost aggregation[J]. Mathematical Biosciences and Engineering, 2021, 18(4): 3215-3226.

[36] 李岩,王春媛,王益涵等.超像素分割约束的自适应SAD与Census融合立体匹配算 法[J].光学技术,2022,48(04):478-485.

[37] Zhang Z H, Wang X F, Yu J W, et al. Disparity refinement based on least square support vector machine for stereo matching[J]. Signal Image and Video Processing, 2022, 16(8): 2141-2148.

[38] 向玺蒙,王竞雪.多尺度引导滤波加权聚合的立体匹配算法[J].测绘科学,2023,48(02):131-139.

[39] Cheng J. Three-dimensional reconstruction of binocular stereo vision based on improved SURF algorithm and KD-Tree matching algorithm[J]. Journal of digital information management, 2015, 13(6):462-466.

[40] Yao L J, Hu D, Yang Z D, et al. Depth recovery for unstructured farmland road image using an improved SIFT algorithm[J]. Int J Agric & Biol Eng, 2019, 12(4): 141-147.

[41] Cai Z, Ou Y, Ling Y, et al. Feature detection and matching with linear adjustment and adaptive thresholding[J]. IEEE Access, 2020, 8:189735-189746.

[42] 潘峰,沈建新,秦顺等.融合GMS的ORB特征点提取与匹配算法[J].计算机工程与设计,2022,43(08):2244-2251.

[43] Muquit M A, Shibahara T, Aoki T. A high-accuracy passive 3D measurement system using phase-based image matching[J]. IEICE Transactions on fundamentals of electronics, communications and computer sciences, 2006, 89(3): 686-697.

[44] 马宁,门宇博,门朝光,等.基于扩展相位相关的小基高比立体匹配方法[J].电子学报, 2017, 45(8): 1827-1835.

[45] 蔡超,刘文波,郑祥爱等.基于多尺度分析的快速相位立体匹配[J].半导体光电,2020,41(06):870-874+878.

[46] Karnaukhov V N, Kober V I, Mozerov M G, et al. Robust Stereo Matching Using Phase Features Based on the Walsh–Hadamard Transform[J]. Journal of Communications Technology and Electronics, 2021, 66(12):1438-1443.

[47] 唐笑虎,胡丹,刘凯.一种极线近似的双目结构光相位立体匹配方法[J].强激光与粒子束,2022,34(11):16-21.

[48] 尹晨阳,职恒辉,李慧斌.基于深度学习的双目立体匹配方法综述[J].计算机工程,2022,48(10):1-12.

[49] Zbontar J, LeCun Y. Stereo matching by training a convolutional neural network to compare image patches [J]. Journal of Machine Learning Research, 2016, 17(1):2287-2318.

[50] Mayer N, Ilg E, Hausser P, et al. A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition.2016: 4040-4048.

[51] Guo X Y, Yang K, Yang W K, et al. Group-wise correlation stereo network[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019:3268-3277.

[52] 杨晓立,徐玉华,叶乐佳等.双目立体视觉研究进展与应用[J].激光与光电子学进展,2023,60(08):180-196.

[53] Guo C, Chen D, Huang Z. Learning Efficient Stereo Matching Network with Depth Discontinuity Aware Super-Resolution[J]. IEEE Access, 2019,7:159712-159723.

[54] Scharstein D, Szeliski R. A Taxonomy and Evaluation of Dense Two-Frame Stereo Correspondence Algorithms[J]. International Journal of Computer Vision, 2002, 47(1):7-42.

[55] 孙延坤,李彩林,王佳文等.融合绝对误差和与Census变换的双目立体图像匹配算法[J].科学技术与工程,2020,20(29):12035-12041.

[56] Lv C, Li J, Kou Q, et al. Stereo matching algorithm based on HSV color space and improved Census transform[J]. Mathematical Problems in Engineering, 2021, 2021:1-17.

[57] 蒋文萍,汪凌阳,韩文超等.基于改进Census变换的自适应局部立体匹配[J].电子测量技术,2022,45(13):82-87.

[58] Hou Y, Liu C, An B, et al. Stereo matching algorithmbased on improved Census transform and texture filtering[J]. Optik, 2022,249:1-9.

[59] 乔景慧,韩玉明,张啸涵.基于改进Census变换的鲁棒立体匹配算法[J].计量学报,2023,44(05):694-700.

[60] 胡志新,梅紫俊,王涛等.基于自适应窗口和改进Census变换的半全局立体匹配算法[J].电光与控制,2023,30(03):33-37.

[61] Mei X, Sun X, Zhou M, et al. On building an accurate stereo matching system on graphics hardware[C]//2011 IEEE International Conference on Computer Vision Workshops (ICCV Workshops). 2011: 467-474.