摘要
立体匹配一直以来都是双目视觉领域中研究的重点和难点。针对现有立体匹配算法边缘保持性差、匹配精度低等问题,提出了一种二次加权引导滤波融合双色彩空间模型的立体匹配算法(Secondary Weighted Guided Filtering fusion double color model,SWGF)。首先在代价计算阶段提出了一种双色彩空间模型,从两个颜色空间进行像素颜色匹配代价计算,增强像素在低纹理区域的特征;然后在代价聚合阶段基于HSV颜色空间利用不同窗口中像素纹理不同加入一个边缘保持项,从而使正则化参数进行自适应调整。在一次引导滤波之后,我们使用Census变换得到的汉明距离和初始视差完成一次代价更新,再次对其进行代价聚合,随后计算视差并对视差进行左右一致性检测、空洞填充和加权中值滤波等优化,最后获得视差图。本文算法在Middlebury测试平台上测试结果表明SWGF算法误匹配率仅为4.61%,可以大幅提升立体匹配的精度,同时增强其边缘保持性。
Stereo matching has always been the fo cus and difficulty in the field of binocular vision.To solve the problems of poor edge retention and low matchi ng accuracy of existing stereo matching algorithms,a stereo matching algorithm based on secondary weighted guided filtering fusion double color model(SWGF)is proposed.Firstly,a dual color space model is proposed in the cost calculation stage,which calculates the pixel color matching cost from two color spaces and enhances the characteristics of pixels in low texture area Then,in the cost aggregation stage,based on HSV color space,an edge preserving term is add ed by using different pixel textures in different windows,so that the regulariz ation parameters can be adjusted adaptively.After a guided filtering,we use Ha mming distance and initial disparity obtained from Census transform to complete a cost update,and then aggregate the cost.Then we calculate the disparity and optimize the disparity by left-right consistency detection,hole filling and we i ghted median filtering.Finally,we get the disparity map.The test results of t his algorithm on Middlebury test platform show that the mismatching rate of SWGF algorithm is only 4.61%,which can greatly improve the accuracy of stereo match ing and enhance its edge retention.
作者
刘徐秀
闫传为
贾晓芬
赵佰亭
LIU Xu-xiu;YAN Chuan-wei;JIA Xiacrfen;ZHAO Bai-ting(School of electrical and Information Engineering,Anhui University of Science and Technology.Huainan 232001,China)
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2020年第8期842-849,共8页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61501006)
国家重点研发计划专项资助(2016YFC0600908)
博士后基金国家面上项目(2016M592035)
安徽省重大科技专项项目(16030901012)
安徽省高校自然科学研究重大资助项目(KJ2018ZD008)
安徽省高校优秀青年人才重点项目(gxyqZD2018036)
安徽省博士后研究人员科研活动经费资助项目(2017B173)
安徽高校协同创新项目(GXXT-2019-048)
安徽理工大学2019年研究生创新基金项目(2019CX2048)资助项目。
关键词
双目视觉
立体匹配
双色彩空间模型
二次引导滤波
binocular vision
stereo matching
double color space model
secondary guided filtering
