基于半全局优化的多视影像匹配方法与应用

发布时间：2018-04-06 05:09

本文选题：多视影像匹配　切入点：半全局约束立体影像匹配　出处：《中南大学》2013年硕士论文

【摘要】：覆盖同一区域的大重叠度影像为影像匹配提供大量的冗余信息,因此多视影像匹配成为数字摄影测量的研究热点。多视影像具有重叠度高的特点,因此,如何在匹配过程中充分利用多视影像的冗余信息,是多视影像匹配的关键。针对影像匹配可靠性问题,本文提出两种基于半全局优化的多视影像匹配方法,在获取密集匹配点云的基础上,对三维重建进行了研究。本文主要研究内容如下： (1)针对断裂遮挡、纹理单一及纹理缺乏等匹配困难区域,本文提出一种基于像方串点的半全局多视影像匹配方法。通过对所有已校正的立体像对进行半全局约束立体影像匹配,根据所有立体像对得到的同名点通过传递追踪的方法实现多视影像串点,然后利用多片前方交会迭代优化实现匹配结果在物方的融合,形成一个整体的匹配结果。多视影像匹配具有较大的匹配冗余,可以提高匹配的可靠性,同时多片前方交会迭代优化有利于提高交会精度,为后续三维建模提供密集可靠的点云。 (2)针对单立体匹配模式在遮挡区域容易产生歧义性匹配,且在纹理单一及纹理缺乏区域容易产生“多义性”匹配问题,本文提出一种基于物方几何约束的半全局多视影像匹配方法。根据影像的成像关系,在物方几何约束下引导多视影像同时进行匹配,通过半全局优化方法减少错误匹配。采用由粗到精的金字塔影像匹配策略,利用低分辨率影像匹配结果约束高分辨率影像匹配,实现匹配传播,减少由于匹配模糊导致的错误匹配,同时有利于减少内存消耗以及降低计算复杂度。 (3)在获取密集可靠的点云基础上,利用Possion表面重建重构场景几何拓扑结构,并将对应影像的纹理映射到三维模型上,获取具有真实感的场景模型。本文提出两种基于半全局优化的多视影像匹配方法,其理论、算法和有关试验使用Visual Studio2010实现。有关试验结果证明本文方法能为三维建模提供密集可靠的三维点,为高精度三维重建提供一条可靠的途径。
[Abstract]:Large overlap images covering the same area provide a lot of redundant information for image matching, so multi-view image matching has become a hot topic in digital photogrammetry.Multi-view image has the characteristics of high overlap. Therefore, how to make full use of redundant information in multi-view image matching is the key to multi-view image matching.Aiming at the reliability of image matching, this paper presents two methods of multi-view image matching based on semi-global optimization. Based on the acquisition of dense matching point clouds, 3D reconstruction is studied.The main contents of this paper are as follows:In this paper, a semi-global multi-view image matching method based on image square string points is proposed for difficult areas such as fault occlusion, single texture and lack of texture.By matching all corrected stereo pairs with semi-global constraint stereo image, according to the same name points obtained from all stereo pairs, multi-view image string points are realized by transfer tracing method.Then the multi-slice forward intersection is used to optimize the fusion of the matching results in the object side to form a whole matching result.Multi-view image matching has a large matching redundancy, which can improve the reliability of matching. At the same time, the iterative optimization of multi-slice forward rendezvous is beneficial to improve the rendezvous accuracy and provide a dense and reliable point cloud for the subsequent 3D modeling.(2) in view of the ambiguity matching in the occlusion region of the single stereo matching pattern, and the "polysemy" matching problem in the single texture and the lack of texture region,In this paper, a semi-global multi-view image matching method based on object-square geometric constraints is proposed.According to the imaging relation of the image, the multi-view image can be matched simultaneously under the constraint of object geometry, and the error matching can be reduced by semi-global optimization method.By using the coarse to fine pyramid image matching strategy, the low resolution image matching result is used to constrain the high resolution image matching, and the matching propagation is realized, and the error matching caused by the matching fuzzy is reduced.At the same time, it can reduce memory consumption and computational complexity.On the basis of obtaining dense and reliable point cloud, the scene geometry topology is reconstructed by Possion surface reconstruction, and the texture of the corresponding image is mapped to the 3D model to obtain the realistic scene model.In this paper, two methods of multi-view image matching based on semi-global optimization are proposed. The theory, algorithm and related experiments are implemented with Visual Studio2010.The experimental results show that this method can provide dense and reliable 3D points for 3D modeling and provide a reliable way for high precision 3D reconstruction.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41;P231.5

【参考文献】