【作者】 王美珍；

【导师】 刘学军；

【作者基本信息】 南京师范大学 ， 地图学与地理信息系统， 2011， 博士

【摘要】 测量几乎是各个领域都涉及到的必不可少的工作之一。图像量测以其非接触性、普适性、实时性、可重复性等特点成为地物对象几何量测的重要手段之一。当前图像获取的方式呈现多元化趋势,并且也无法得到摄影设备参数。另外,不同来源的图像之间摄影基线无法事先控制,这些都加大了双目视觉应用中图像匹配、相机自标定等过程的难度。因此,必须破除传统摄影测量来自“双眼视觉”的束缚。基于单幅图像的量测技术仅利用单幅图像中蕴含的几何特征,避免了立体视觉中图像之间匹配、相机自标定等经典难题,成为图像量测发展的重要趋势。本文以单幅未标定近景图像为研究对象,旨在提取单幅图像中地物目标的几何尺寸。通过研究取得以下成果：(1)面向单幅图像几何量测的先验知识分类与先验知识库架构。在总结现有单幅图像几何量测方法中对已知信息运用的基础上,根据先验知识在单幅图像几何量测过程中所起的作用,将先验知识分为：几何约束信息、度量参考信息以及语义约束信息。在系统梳理地理场景中的各类先验知识的基础上,结合人类对地理场景的认知特点,以两类数据表——对象分类层次表和对象度量信息表描述先验知识,前者主要表示实体对象所属的类别,方便地理对象快速检索,后者主要描述地理对象的度量参考信息。在此基础上,本文进而实现了先验知识库的架构。(2)针对地理场景中矩形对象,提出基于交比的几何量测方法。考虑到交比为射影变换不变量,本文结合灭点、灭线得出以下三个结论：①已知一条直线上某段线段的长度以及直线的灭点,则可获得直线上任意两点之间的距离。②已知平面内一条参考线段的长度及所在平面的灭线,即可获得参考平面中与参考线段平行的直线上任意两点之间的距离。③已知平面上一矩形及两条边长,则可获得平面上任意两点之间的距离。在距离量测的基础上,指出了其他几何信息的计算方法以及三维空间上(主要指垂直平面上、平行平面上)地物几何信息的量测方法,测试表明了算法的可用性。(3)考虑地理场景另外一类常见几何特征圆形,在上文基于交比的三个结论的基础上,提出了三种距离量测方法：基于圆形直径上交比的方法、基于圆形外接矩形的方法、基于圆形两条垂直直径的方法。通过实验证明：基于圆形直径上交比的方法量测结果对误差具有很强的敏感性,测量结果各个方向精度差异大；而后两种方法量测结果则较为稳定、量测结果精度有所提高。(4)在总结现有几何量测相关算法的基础上,对基于相同特征的不同灭点、灭线、几何量测方法进行对比测试分析。通过测试分析得出,在同等几何特征提取精度下,优先选用基于平行线的灭点计算方法；针对包含同心圆的图像优先选用基于圆形割线的灭线计算方法；在包含矩形的场景中可选用本文提出的基于复合交比的几何量测方法或直接线性变换；而在包含圆形的场景中,可选用本文提出的基于圆形外接矩形方法。(5)在上文总结的算法优选策略基础上,针对一幅图像中包含不同的参考对象的情况,通过对包含多个矩形和圆形的多幅图像几何量测算法进行测试,结果分析表明,①当参考对象位于图像中心时,待测线段的整体量测精度较高；②针对某一待测对象,就近选择参考对象可获得较高的量测精度。针对可选用多个参考对象的情况,采用灭线优化算法对几何量测过程进行优化,采用加权平均方法对几何量测结果进行优化,测试表明了策略的正确性,尤其在优选参考对象之后,几何量测结果精度更高。(6)基于论文中所提出的方法,设计并实现了面向结构化场景单幅图像几何量测的原型系统。该系统具有几何特征提取、先验知识入库、检索、针对不同几何特征的单幅图像距离量测、其他几何信息计算等功能。本研究充分利用单幅图像本身所蕴含的线索,挖掘图像中地物目标的几何尺寸信息,为GIS空间数据获取提供一条新途径;本研究的数据获取方便、量测方法方便快捷,使得公众能够体验到GIS的便利性,进而推动了GIS社会化发展；同时,本研究可为公安、交通、数字城管、古建筑物三维重建、应急救援等领域提供决策支持。更多还原

【Abstract】 Geometric Measurement is one of the essential jobs almost involving every field. As characterized by its non-contact and universality, real-time and repeatability, image measuring gradually becomes one of the important methods for man-made features measurement. Currently, ways for image acquisition present a diversified tendency without getting innerior or extrior camera parameters, and the free of beforehand control of photography baseline between images from different sources greatly increase the difficulty of the stereoscopic vision applications, such as feature point matching, camera self-calibration and other processes. Therefore, it is necessary to break through the shackles of "stereoscopic vision" in traditional photogrammetry. The measurement technology based on single image just takes use of geometric features in single image to avoid classical problems in stereo vision such as eature point matching and camera self-calibration, appearing an important trend for image measuring development.By using photographs obtained from non-metric equipments, such as ordinary digital camera, video camera, monitoring camera etc and single image acquired from the internet, this thesis constructs a priori knowledge base supporting geometric measurement, researches geometric measurement methods for single image, builds a measurement platform based on geometric information from single image, and excavates geometric information implied in images. It not only develops a new way for GIS spatialdata acquisition, but also provides decision supports for criminal investigation, traffic accidents, digital urban management,3D reconstruction of ancient building, emergency rescue and other fields. The main contributions are as follows:(1) Classified priori knowledge and constructed a priori knowledge base. First of all, according to the role of prior knowledge through the process of single image geometric measurement, priori knowledge has been divided into three classfications:geometric constrained information, measuring referenced information and semantic constrained information. On that basis, all classifications of priori knowledge in geographical scenes have been combed, data tables and field information in the priori knowledge base have been designed according to the characteristics of prior knowledge, and finally the priori knowledge mainly including measuring referenced information has been constructed. (2) Proposed the geometric measurement method based on rectangle. Considering the invariance of cross ratio in projective transformation and the characteristics of vanishing point and vanishing line, three following conclusions can be drew. Conclusion one:distance between any two points on one line can be calculated on the condition of knowing the length of that line plus a vanishing point on it. Conclusion two:once known the length of a referenced line segment and a vanishing line in a plane, distance between any two points on a line which is parallel with the referenced line segment in the same plane. Conclusion three:if a rectangle and the length of its two sides on a plane has been given, then distance between any two points in this plane can be obtained. Combined with the above conclusions, other geometry information and geometry information in three-dimensional space can be acquired further.(3) Proposed the geometric measurement method baseds on circle. For diameters of a circle owning the same properties in all directions, three methods have been proposed, which are respectively based on cross ratio of circle diameter, based on circum-rectangle of a circle and based on DLT. However, as proofed by experiments, results of the first method which is based on cross ratio of circle diameter are different in all directions for the impact of vanishing line and so on, while the other two methods do not have this problem. This means based on the circum-rectangle of a circle can be applied in geometry measurement of a plane containing the following geometric features:two circles(concentric circles, separation, intersection and tangent), an independent circle with two known diameters, a circle with one known diameter and a line parallelled with the known diameter, and a circle with one known diameter and a right angle.(4) Summarized optimization strategy for geometric measurement accuracy under multiple constraint conditions. Through tests and analyses, it is prior to choose the method based on vanishing points of parallel lines in the same circumstances; it is better to use the way based on rectangle and the one based on circum-rectangle of a circle in the process of geometric measuring; while multiple known referenced geometric characteristics existing, the rule for choose of referenced objects is mainly in line with spatial relationships between unmeasured objects and referenced objects. In a variety of situations, to optimize the geometric measuring process, the vanishing line optimization algorithm is usually adopted, and to improve results of the geometric measuring, the weighted average method is often applied. The validity of the suggested strategies has got proofed through tests, and correspondingly the precision for geometric measuring results has been improved after referenced objects having been optimized. (5) Designed and implemented the prototype system of single image geometric measurement catering to structural scenes. This prototype concludes following functions: extraction of geometric features, inbound and retrieval of prior knowledge, distance measurement for single image with different geometric characteristics, and calculations for other geometric information etc.This dissertation took full advantage of the clues in single image and obtained the geometric infromation of the objects in the image. Firstly, It offers a new way for GIS spatial data acquisition. Secondly, it is relatively easier to extract geometric information from images, and it makes that GIS can be accessed by common people, thus boosts the GIS social development. Finally, It can be employed in public security, traffic, forensic science,3D reconstruction for ancient buildings and so on.更多还原