基于近景图像序列的建筑物三维模型重建研究

发布时间：2018-01-03 18:40

本文关键词：基于近景图像序列的建筑物三维模型重建研究　出处：《南京师范大学》2013年博士论文　论文类型：学位论文

【摘要】：建筑物的数据获取与三维建模是数字城市、智慧城市建设的重要内容,基于近景图像序列的三维建模方法以普通数码相机为图像采集设备,成本低、效率高、劳动强度低,能够反映建筑物表面的几何细节且具有逼真的纹理信息,为快速、准确、真实地再现三维城市信息提供了有效的途径。然而,由于地面近景图像往往采用宽基线、大交角的摄影方式,导致遮挡严重,图像间透视畸变较大,加上建筑物纹理单一,重复多等问题,给图像信息的提取和匹配带来困难,建筑物三维重建的精度和自动化程度还有待提高。在图像信息中,点特征定位精确,可以有效恢复二维图像与三维几何之间的映射关系,而直线特征是建筑物的主要轮廓特征,对整个建筑物结构起控制作用。为充分发挥点特征和直线特征的优势,本文就基于近景图像序列点、线特征的建筑三维建模过程中涉及的几个关键技术问题展开研究,主要工作和研究成果包括以下几个方面： (1)针对由于近景图像视角变化大、建筑物纹理单一等导致的特征点匹配困难问题,重点研究对视角变化保持不变的局部仿射不变特征,在分析了目前局部仿射不变特征检测方法优缺点的基础上,提出一种多特征优势互补的特征点检测与匹配方法,得到精确度较高和数量较多的匹配点对,为后续相机参数估计奠定了较好的基础。 (2)针对相机参数估计的鲁棒性问题,以相机参数与三维点坐标同步解算的光束平差优化原理为基础,根据近景图像特征点透视畸变导致误差分布呈各向异性的特点,以目标函数的构建为切入点,提出了一种顾及特征点误差各向异性的相机参数及三维点坐标解算方法。通过对比实验,验证了本文方法具有较高精度和较好的鲁棒性。 (3)鉴于直线特征对建筑物三维几何结构的重要性,设计了直线特征的提取、匹配与重建策略。在进行直线特征提取的基础上,针对直线断裂、提取不完整导致的匹配困难,提出了一种多约束的直线特征匹配方法,该方法的特点在于利用点、线之间的局部仿射不变进行候选匹配集合的搜索,提高了匹配直线搜索的正确率,结合极线约束和直线的角度及支持域灰度的相似性测度进行直线特征匹配,实验结果表明该方法具有更高的正确匹配率。将本文的匹配方法应用于直线特征的三维重建,基于面-面相交确定空间直线的原理,以迭代优化实现了直线特征的三维重建,较好地恢复了三维直线的位置及平行、垂直等关系。 (4)结合本文提出的方法和前人的研究成果,实现了综合点、线特征的建筑物三维模型重建。在分析了当前建筑物模型表达方法的基础上,实现了基于约束Delaunay三角剖分的建筑物三维几何模型重建,并根据相机成像原理完成了二维图像到三维模型的纹理映射,通过实例验证了本文方法的有效性。
[Abstract]:Building data acquisition and 3D modeling is an important part of the construction of digital city and intelligent city. The 3D modeling method based on close-range image sequence takes the common digital camera as the image acquisition equipment, which has the advantages of low cost and high efficiency. Low labor intensity, can reflect the geometric details of the building surface and have realistic texture information, which provides an effective way to reproduce 3D urban information quickly, accurately and truthfully. Because the ground close-range image often uses the wide baseline, the big intersection angle photography, causes the occlusion to be serious, the image perspective distortion is big, plus the building texture is single, repeats and so on the question and so on. It is difficult to extract and match the image information, and the accuracy and automation of 3D reconstruction of buildings need to be improved. In the image information, the point feature location is accurate. The mapping relationship between 2D images and 3D geometry can be effectively restored, and linear features are the main contour features of buildings. In order to give full play to the advantages of point features and linear features, this paper is based on the close-range image sequence points. In the process of building 3D modeling with line features, several key technical issues are studied. The main work and research results include the following aspects: 1) aiming at the difficulty of matching feature points caused by the large change of view angle and the single texture of building, this paper focuses on the local affine invariant feature which keeps the change of view angle invariant. Based on the analysis of the advantages and disadvantages of the existing local affine invariant feature detection methods, a multi-feature complementary feature point detection and matching method is proposed to obtain a high accuracy and a large number of matching points. It lays a good foundation for the subsequent camera parameter estimation. 2) aiming at the robustness of camera parameter estimation, the principle of beam adjustment optimization is based on the synchronous calculation of camera parameters and 3D coordinates. According to the anisotropy of the error distribution caused by the perspective distortion of feature points in close-shot images, the construction of objective function is taken as the starting point. In this paper, a camera parameter and 3D point coordinate calculation method considering the anisotropy of feature point error is proposed. The comparison experiment shows that the proposed method has higher accuracy and better robustness. In view of the importance of straight line feature to 3D geometric structure of building, the extraction, matching and reconstruction strategy of straight line feature are designed. On the basis of line feature extraction, straight line fracture is aimed at. In this paper, a multi-constraint line feature matching method is proposed. The feature of this method is to search candidate matching sets by using the local affine invariance between points and lines. The accuracy of matching line search is improved, and line feature matching is carried out by combining the pole line constraint, the angle of the line and the similarity measure of the support domain gray level. The experimental results show that this method has a higher correct matching rate. This method is applied to 3D reconstruction of line features, and the principle of plane-plane intersection to determine spatial straight line is presented. The 3D reconstruction of line features is realized by iterative optimization, and the position, parallel and vertical relationships of 3D straight lines are restored. 4) based on the method proposed in this paper and the previous research results, the 3D model reconstruction of buildings with synthetic points and line features is realized. The 3D geometric model reconstruction of buildings based on constrained Delaunay triangulation is realized, and the texture mapping from 2D image to 3D model is completed according to the principle of camera imaging. The effectiveness of this method is verified by an example.
【学位授予单位】：南京师范大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：P234.1;TU19

【参考文献】