基于连续对称视差计算的多视图立体匹配  

Multi-view three-dimensional reconstruction using continuous symmetric disparity

作　　者：朱文峤[1] 刁常宇[2] 许端清[1] 鲁东明[1] 

机构地区：[1]浙江大学计算机科学与技术学院,浙江杭州310027 [2]浙江大学文化遗产研究院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2014年第1期85-91,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家"十二五"科技支撑计划资助项目(2012BAH03F02);浙江省重点创新团队建设计划资助项目(2010R50040);中央高校基本科研业务费专项资金资助项目(012X22X003-3);国家社会科学基金重大招标项目(12&ZD232)

摘　　要：为了解决基于非凸能量最小化的多视图三维重建方法容易陷入到局部解的问题,提出鲁棒的基于连续对称视差的多视图立体匹配算法.采用连续对称优化算法对基于局部窗口的算法产生的视差进行优化,在能量函数中引入左、右一致性约束,为了提升视差图的准确度,采用近邻的图像和视差图迭代地对视差进行改进.每一个视差图对应一个深度图,因为生成的视差图有较好的质量,融合的过程可以采用简单的算法.Middlebury的测试图像和现实中的一些场景的重建的实验效果证明了该算法的有效性.A robust symmetric continuous disparity based multi-view stereo algorithm was proposed in order to avoid the local optimum caused by the non-convex energy minimization based algorithm. The sym- metric continuous method was applied to improve the disparity maps estimated by windows based algo- rithm. Left and right consistency soft constraint was used for more robust disparity estimation. Neighbor images and disparity maps were further used in order to alleviate the limitation of the continuous method which easily falls into a local minimum. The aforementioned steps were iterated until a high quality dispar- ity map. Simple depth fusion algorithms can be applied because of the quality of the disparity map. The proposed algorithm was tested with both the data sets from Middlebury site and the real world objects, which showed the effectiveness of the algorithm.

关 键 词：多视图立体匹配 连续优化法 连续对称视差计算 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]