李宝平, 王宏涛.基于两阶段自适应优化的双目立体匹配算法[J].河南理工大学学报（自然科学版）,2020,39(1):125-132.

LI B P, WANG H T.Binocular stereo matching algorithm based on two-stage adaptive optimization[J].Journal of Henan Polytechnic University(Natural Science) ,2020,39(1):125-132.

基于两阶段自适应优化的双目立体匹配算法

李宝平1, 王宏涛2

1.河南理工大学物理与电子信息学院，河南焦作 454000;2.河南理工大学测绘与国土信息工程学院，河南焦作 454000

摘要:针对局部立体匹配算法在弱纹理区域匹配准确度低的问题，提出一种在代价初始化和代价聚合两阶段自适应优化的立体匹配算法。首先，在代价初始化阶段，通过双阈值线性约束条件构造像素点的十字支撑窗口，并依据十字支撑最短臂长构造自适应函数融合AD及Census 特征测度；其次，在代价聚合阶段，利用十字支撑水平扩展及形态学处理的方法构造自适应滤波窗口，并通过区域滤波实现代价聚合；最后，通过视差选择及视差优化得到最终的视差图像。 Middlebury平台测试结果表明：该算法与传统的AD-Census融合立体匹配算法相比，图像集的整体匹配误差减小了 3.95%，在非遮挡区域的匹配误差减小2.17% ；与传统区域滤波立体匹配算法相比，该算法在弱纹理区域可以取得更好的立体匹配效果。

关键词:双目立体匹配算法;两阶段自适应优化;区域滤波;十字支撑

doi:10.16186/j.cnki.1673-9787.2020.1.16

基金项目:河南省科技攻关计划资助项目（172102310350 ）；河南省自然科学基金面上项目（182300410115 ）；河南理工大学博士基金 资助项目（B2017-55 ）

收稿日期:2019/03/18

修回日期:2019/04/29

出版日期:2020/01/15

Binocular stereo matching algorithm based on two-stage adaptive optimization

LI Baoping1, WANG Hongtao2

1.School of Physics & Electronic Information Engineering， Henan Polytechnic University， Jiaozuo  454000 ， Henan， China;2.School of Surveying and Land Information Engineering， Henan Polytechnic University， Jiaoz'uo  454000 ， Henan， China

Abstract:Aiming at the low accuracy problem of local stereo matching algorithm in weak texture region， a two-stage adaptive stereo matching algorithm based on cost initialization and cost aggregation was proposed. In cost initialization stage， the cross support window of pixel was constructed through double thresholds linear constraint ，and the adaptive function was constructed according to the shortest arm length to fuse AD and Census feature measures； in cost aggregation stage， an adaptive filter window was constructed by using cross support extension and morphological processing， and cost aggregation was realized by region filtering ； finally， the final parallax image could be obtained by parallax selection and refinement. The results of experiments on the Middlebury platform indicated that， compared with the traditional AD-Census fusion stereo matching algorithm， the overall matching error of the image set decreased by 3.95% ， and the matching error in the non-occlusion area decreased by 2. 17 % . Compared with the traditional filtering stereo matching algorithm， the proposed algorithm achieved better stereo matching effect in the weak texture region.

Key words:binocular stereo matching algorithm;two-stage adaptive optimization;region filtering;cross support

基于两阶段自适应优化的双目立体匹配算法_李宝平.pdf