拓扑几何约束的多视影像快速匹配及三维重建方法研究

Multi-view image matching and three-dimensional reconstruction enable automatic and efficient acquisition of geometric structures. In order to obtain the rich details of the environment, the traditional basis of wide-based imaging is not preferred. On the contrary, photography in the design of short baseline, multi-angle and high overlap will produce massive data, making explosive increment of computation with the number of images, thus needs new methods and techniques. Hence, it has become a new research hotspot in the field of digital photogrammetry and computer vision to solve the blind matching problem of redundant images. This project will focus on the novel methods of multi-view image matching and three-dimensional reconstruction based on topological geometric features. The detailed research topics include: (1) A fuzzy method will be approached to adaptively identify a minimally required number of image set, considering integrated topological geometric features. (2) An optimization method will be conducted to filter out redundant adjacencies from a dense adjacency matrix for guiding image matching. (3) The skeleton of the image topology will be extracted, based on which the large set of images can be divided into subsets for parallel 3D reconstruction. (4) The experiments to test the feasibility and correction of our methods above will be done using different kinds of image sets from multiple platforms. Our achievements are promising to provide a new framework for intelligent, fully automatic, near-real-time, and high-precision three-dimensional information acquisition, and to provide possible solutions for large-scale deformation monitoring, disaster emergency management, mining, and digital city construction applications.

多视影像匹配及三维重建是自动获取三维信息的有效手段，但为了获取被摄环境中丰富的细节信息，传统基于宽基线的摄影方式常常无法满足需求。相反，短基线、多角度、高重叠的摄影方式会产生大量影像，且处理耗时随影像数呈二次方爆炸式增长，效率极低，需研究新的方法和技术。因此，解决冗余影像的盲匹配问题成为数字摄影测量与计算机视觉学术界一个新的研究热点。本项目针对这一学术前沿和实际应用需求，研究拓扑几何约束的多视影像快速匹配及三维重建方法。包括：（1）顾及拓扑几何多要素模糊约束的最小影像集自适应优选；（2）稀疏邻接矩阵提取及其约束影像匹配；（3）顾及影像拓扑骨架结构的影像空间智能划分与分层三维重建；（4）并以不同场景下的多视影像进行试验和算法验证。有望为大范围的形变监测、灾害应急、矿采评估、数字城市等综合应用与决策分析提供智能化、全自动、准实时和高精度三维信息获取模式。

多视影像匹配及三维重建是自动获取三维信息的有效手段，但为了获取被摄环境中丰富的细节信息，传统基于宽基线的摄影方式常常无法满足需求。相反，短基线、多角度、高重叠的摄影方式会产生大量影像，且处理耗时随影像数呈二次方爆炸式增长，效率极低，需研究新的方法和技术。因此，解决冗余影像的盲匹配问题成为数字摄影测量与计算机视觉学术界一个新的研究热点。本项目针对这一学术前沿和实际应用需求，研究拓扑几何约束的多视影像快速匹配及三维重建方法。包括：（1）顾及拓扑几何多要素模糊约束的最小影像集自适应优选；（2）稀疏邻接矩阵提取及其约束影像匹配；（3）顾及影像拓扑骨架结构的影像空间智能划分与分层三维重建；（4）并以不同场景下的多视影像进行试验和算法验证。基于课题研究，从近景摄影到航空摄影，从静态摄影到视频摄影系统分析了拓扑几何要素对多视影像匹配精度的影响，提出了两种拓扑几何要素约束的影像优选模型；据此，发展了多要素约束的稀疏邻接矩阵提取及其约束影像匹配模型，提高了多视影像匹配效率和精度；针对大场景影像重建问题，设计了分区重建联合多站点云配准策略， 并提出了顾及法线和多尺度稀疏特征约束的点云全局配准方法以及顾及形态不变特征匹配的局部配准方法，解决了复杂大场景多站、多时相点云配准问题；研究成果已在地震灾场应急测量与损毁评估测图、矿区开采沉陷监测、矿采三维态势评估、数字城市建模等综合应用与决策分析提供了智能化、全自动、准实时和高精度三维信息获取模式。