结合超像素分割和多尺度的快速半全局立体匹配算法  

A Fast Semi-global Stereo Matching Algorithm Combining Superpixel Segmentation and Multiscale

作　　者：刘子濠 郭龙源 王琼[1,2] 赵业涛 唐峰 曾毅 LIU Zihao;GUO Longyuan;WANG Qiong;ZHAO Yetao;TANG Feng;ZENG Yi(School of Information Science and Engineering,Hunan Institute of Science and Technology,Yueyang 414006,China;Machine Vision&Artificial Intelligence Research Center,Hunan Institute of Science and Technology,Yueyang 414006,China)

机构地区：[1]湖南理工学院信息科学与工程学院,湖南岳阳414006 [2]湖南理工学院机器视觉与人工智能研究中心,湖南岳阳414006

出　　处：《成都工业学院学报》2023年第4期53-59,共7页Journal of Chengdu Technological University

基　　金：湖南省自然科学基金资助项目(2020JJ5218);湖南省教育厅项目(19A200);湖南省三维重建与智能应用技术工程研究中心项目(2019-430602-73-03-006049)。

摘　　要：半全局立体匹配算法采用最大视差范围进行搜索,有较大冗余。为提高算法的效率,提出一种结合超像素分割和多尺度的快速半全局立体匹配算法。首先获得原图1/4大小的视差图;其次,结合原图像超像素分割结果,计算每个分割块的平均视差,并在平均视差基础上设置上下阈值,作为该分割区域所有点的视差搜索范围。这样,既减少了视差搜索范围,也保证了搜索结果的精度;同时,为降低原始视差图误匹配率,在计算弱纹理区域的聚合值时,叠加了下采样层对应区域的聚合值。弱纹理区域是根据超像素分割区域的灰度直方图确定的;最后,在已缩小的搜索范围内,根据代价最小原则获得最终的视差。实验结果表明:该方法在保证视差精度的前提下,计算复杂度大幅降低,计算时间缩短为原来的1/4。The semi-global stereo matching algorithms has lots of redundancy for searching with the maximum disparity.To improve the efficiency of the algorithm,a fast semi-global stereo matching algorithm that combines super-pixel segmentation and multi-scale information was proposed.Firstly,the disparity map of 1/4 of the original image was obtained;Secondly,combined with the result of the super-pixel segmentation of the original image,the average disparity of each segment area was calculated.Then,the disparity search range of all points in the segment were set based on the average disparity and the upper and lower thresholds.In this way,not only the disparity search range was reduced,but also the accuracy of the search results was guaranteed;At the same time,in order to reduce the error matching rate of the original disparity image,the aggregate value of the corresponding area of the down-sampling layer was superimposed,when calculating the aggregation value of the weak texture.The weak texture areas were determined according to the grayscale histogram of the super-pixel segmentation.Finally,the final disparity was obtained according to the principle of minimum cost in the narrowed search range.The experimental results show that this method makes the computational complexity greatly reduced and the computation time is shortened to 1/4 of the original on the premise of ensuring the disparity accuracy.

关 键 词：多尺度 超像素分割 立体匹配 视差搜索范围 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]