基于单幅建筑物图像的三维信息提取  

Three-dimensional information extraction from a single architecture image

作　　者：丁伟利[1] 朱枫[1] 郝颖明[1] 

机构地区：[1]中国科学院沈阳自动化研究所

出　　处：《仪器仪表学报》2008年第9期1965-1971,共7页Chinese Journal of Scientific Instrument

摘　　要：提出并实现了一种利用矩形结构从单幅建筑物图像中提取三维信息的方法。该方法只需要单幅图像和建筑物中易于测量的矩形一边长度作为输入,就可以获得场景中的各种三维信息。本文首先利用空间三方向正交的平行线束求出图像中的消失点,标定相机内参;然后,利用消失点、相机内参和矩形一边长度,对场景中不同位置关系的平面上三维信息的提取方法进行了研究,给出了2个一般性的结论,同时提出了一种物体高度测量的新方法;最后,利用人机交互的方式提取了室内、外场景的三维信息,并对计算结果的精度进行了分析。实验结果表明,本文方法具有所需条件少、测量信息多的特点,对于三维重建、虚拟现实等领域具有很高的应用价值。An effective approach to extracting three-dimensional information from a single architecture image is presented in this paper. Without any prior knowledge about the camera, we describe how 3D Euclidean information may be extracted from a single architecture image, when only one side length of the rectangle is known. Firstly, the intrinsic parameters of the camera are calibrated by using the three vanishing points of mutually orthogonal directions. Secondly, the algorithms for extracting 3D information are described in addition to two general conclusions. Furthermore, a new approach to height measurement is presented. Finally, the 3D information of the building is extracted with user assistance. The results indicate that the method has many advantages such as requirement of less conditions, large scale of measurement and satisfying resolution.

关 键 词：单视测量 单应矩阵 视觉定位 消失点 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]