文献类型：期刊文献

英文题名：A probabilistic stereo matching model for slanted surface

作者：Fu, Jibin[1]; Wu, Huixin[2]; Zhai, Zhengang[3]

第一作者：傅继彬

通讯作者：Fu, J.

机构：[1] School of Computer and Information Engineering, Henan University of Economics and Law, Zhengzhou, China; [2] School of Information Engineering, North China University of Water Resources and Electric Power, Zhengzhou, China; [3] Zhengzhou Branch, Red Cross Society of China, Zhengzhou, China

第一机构：河南财经政法大学计算机与信息工程学院

通讯机构：[1]School of Computer and Information Engineering, Henan University of Economics and Law, Zhengzhou, China|[1048412]河南财经政法大学计算机与信息工程学院;[10484]河南财经政法大学;

年份：2012

卷号：4

期号：15

起止页码：226-234

外文期刊名：International Journal of Advancements in Computing Technology

收录：EI(收录号：20123715422414);Scopus(收录号：2-s2.0-84865710975)

语种：英文

外文关键词：Binoculars - Pixels - Textures

摘要：The frontal-parallel surface assumption is used implicitly or explicitly by many conventional stereo matching algorithms. However, the assumption may not be true for many slanted planes in real world. Consequently, it results in systematic error in stereo matching. In this paper, a new probabilistic model for handling slanted planes is presented. A coarse stereo matching step is used to estimate an initial disparity maps for the images. The monocular component and binocular component of pixel is used to refine probability for the slanted plane. The contribution of neighboring pixels to the probability of pixels assigned to the ground is gives a weight to measure. The weights are correlated with monocular information, namely, similarity of color, spatial distance and binocular information, the dissimilar property of pixels. Experimental results of the test bed images with slanted planes, less-textured regions and ground truth demonstrate good performance of the model.