【作者】 武雪玲；

【导师】 杜清运；

【作者基本信息】 武汉大学 ， 地图制图学与地理信息工程， 2009， 博士

【摘要】 因人类的户外活动与地理空间位置息息相关,促使地理信息系统(Geographic Information System, GIS)的迅速发展。从空间信息可视化的视角来看,GIS是对客观地理世界的数字化描述,GIS通过空间分析获得空间信息所反映的规律和趋势,为相关应用提供新的知识,进而辅助工程设计、规划、管理和决策等服务。从空间认知的视角来看,模拟地图和二维电子地图作为地理环境的抽象模型,在形象逼真地表达空间信息方面,存在一定的局限性,而已兴起的虚拟现实(Virtual Reality, VR)技术强调用户在三维虚拟空间中的沉浸感,在一定程度上提高了人机交互性能,进而提升用户的空间认知水平。伴随着移动GIS和可穿戴计算(wearable computing)的出现,GIS被拓展到户外作业,户外增强现实(Outdoor Augmented Reality, Outdoor AR)空间信息表达致力于将虚拟空间信息叠加到真实地理场景上,增强、拓展用户对地理场景的认知。从人机界面的视角来看,目前流行的3DGIS呈现给用户一个可视化的三维虚拟世界,但绝大多数的人机交互限于平面显示器,很难让用户融入其中,户外AR技术为解决该问题提供了一种新的思路。本文首先以户外AR和空间信息可视化表达为理论基础,深入分析两者之间结合的可行性；其次,详细设计户外AR空间信息表达的整体框架并研究其关键技术；最后,重点剖析虚实配准算法、虚实融合和人机交互等技术细节。在研究中,虚实配准是关键点和难点,本文采用两种思路解决该问题：(1)在系统分析室内AR基于标识检测的虚实配准算法基础之上,针对户外AR,提出采用基于混合硬件的虚实配准方法,即使用GPS结合三维电子罗盘获取用户头部位置和姿态数据,该方法具有完备的理论基础,但要面对相机定标这个复杂而繁琐的计算过程；(2)汲取计算机视觉中的算法思想,提出通过建立全局仿射坐标系,解算虚拟物体到真实场景图像的投影坐标,找到获取像素相对深度参数的方法,进而避开相机定标这个复杂过程。论文最后对两种思路的虚实配准方法、虚实融合及人机交互分别进行验证性实践。人获取的75%以上的外界信息来自视觉,图像和图形是承载视觉信息量最大的媒体,而空间信息天然的具有可视化特征。本文针对基于视觉的AR空间信息表达进行集中讨论,在理论层次上,旨在增强人对空间对象及其空间分析结果形象而深层的认知,为地理信息科学的发展开辟新的研究方向,在应用层次上,旨在为户外AR系统提供空间位置相关的信息源,同时,也为空间信息可视化表达提供新的人机界面。本文具体开展了如下研究工作：1.从理论上分析了AR与空间信息认知与表达的关系,这种关系体现在共同的认知基础和移动性,户外AR的提出和移动GIS的出现为两者的结合提供了强大的理论基础和技术方法。从增强空间认知视角出发,两者的结合为空间信息可视化表达提供了新思路和新界面。2.分析并提出了户外AR空间信息表达的基本特征、应用及组成结构。从技术推动的视角,分析了已有的技术原型,为构建和完善户外AR空间信息表达的研究框架提供参考,提出户外AR空间信息表达的研究框架,并讨论了四个关键技术：显示设备、跟踪定位、虚实融合和人机交互。3.提出并实践GPS结合三维电子罗盘的混合虚实配准方法。首先,分析了世界坐标、相机坐标、虚拟物体坐标、像平面坐标和像素坐标等的定义及转换关系,分析提出相机外部参数和内部参数的标定方法；其次,改进已有的室内AR基于标识检测的虚实配准算法,将其中的相对场景坐标用世界坐标替换,进而用与地理位置密切相关的空间信息来增强真实地理场景,设计了一种基于计算机视觉平面标志点的三维虚实配准算法；最后,提出采用GPS结合三维电子罗盘直接获取用户头部位置和姿态数据的基于硬件跟踪设备的虚实配准方法：通过人机交互的方式确定虚拟物体及属性信息在3D欧氏世界坐标系中的位置和姿态,采用GPS结合三维电子罗盘跟踪用户头部位置和姿态,得到从世界坐标到相机坐标的转换矩阵,已知相机内部参数的前提下,将相机坐标转换到2D像平面坐标,实现虚拟物体及属性信息与真实场景图像的合成显示。4.提出并实践基于仿射变换技术的虚实配准方法。近距离观察地理目标时,为了避开位置跟踪和相机定标这个复杂而繁琐的过程,依据Koenderink和vanDoorn、Ullman和Basri于1991年提出的,已知4个非共面的三维空间特征点，那么三维空间中的任意一个点的投影变换都可以用这4个点投影的线性组合来计算的结论,建立一个由世界坐标、相机坐标和虚拟坐标合并而成的全局仿射坐标系。首先,从场景中提取4个特征点,并通过立体视觉的方法计算虚拟物体关键点的全局仿射坐标,结合4个场景特征点投影的线性组合齐次坐标表示,计算得到虚拟物体关键点的投影坐标；其次,通过向量叉乘方法获取像平面的视觉方向,得到像素的相对深度,以实现场景中物体隐藏面消隐；最后,提出2D场景图像中基准特征点的提取算法,并分析了户外虚实信息动态配准误差的产生及减少途径。5.虚实融合和人机交互也是户外AR空间信息表达的重要特点,反应了对显示质量和对“以用户为中心”的自然高效人机交互方式的要求,本文从地理目标建模、虚实融合、人机交互三个方面进行了讨论。更多还原

【Abstract】 Almost all human outdoor activities is closely related to geographical position, thus the development of GIS is more rapidly. From the perspective of spatial information visualization, GIS is a digital description of objective geographical world, and provides relevant applications with new knowledge which comes from patterns and trends reflected from spatial analysis to support engineering design, planning, management and decision-making services; from the perspective of spatial cognition, as abstract models of geographical environment, analog maps and 2D electronic maps have respective limitations in vivid representing spatial information. VR emphasizes user’s sense of immersion in virtual space and improves performance of Human-Computer Interaction to a certain extent. With the emergence of mobile GIS and wearable computing, operating space of GIS is extended to outdoor. Outdoor Augmented Reality spatial information representation is committed to enhance and expand user’s spatial cognition by overlaying virtual geographic objects on real scene; from the perspective of HCI, popular 3DGIS show users a 3D virtual world, but the vast majority of HCI is limited to plane display and is difficult to users to integrate into it. Outdoor AR provides an idea to solve these questions.Based on theories of outdoor AR and spatial information visualization representation, the paper firstly discusses feasibility of combination between the two; secondly, it designs the whole framework and discusses key technologies; thirdly, it focuses on virtual-real registration, virtual-real combination and HCI. In the study, virtual-real registration is critical and difficult. It uses two ideas to solve the problem. Based on analysis existing fiducial detection registration method in indoor AR, it proposes a hybrid hardware registration method which uses GPS and 3D electronic compass to obtain position and pose of user’s head. The method is based on sound theory, but it includes camera calibration, which is a very complicated calculation process. For the purpose of avoiding the complicated camera calibration in the above idea, it set up global affine coordinate system to compute projection coordinates of virtual objects into real scene images. Simultaneously, it brings forward an approach to get pixels relative depth. Finally, it verifies the two registration methods, virtual-real combination and HCI, respectively.Visual information is more than 75% in the whole information obtained from outside world by people. And images and graphics include the largest amount of visual information. Spatial information has natural characteristics of visualization. This paper focuses on vision-based AR spatial information representation. At the theoretical level, it enhances deep understanding of spatial analysis results and explores a new research direction for Geographical Information Science. At the application level, it provides location-related information sources for outdoor AR and becomes a new interface for spatial information visualization representation. The paper comprises five basic components:Firstly, it analyzes relation between AR and spatial information cognition and representation which embodies in the common basic cognition and mobility. Then the emergence of outdoor AR and mobile GIS provides combination of outdoor AR and spatial information representation with powerful theoretical foundation and technical channel. From the perspective of enhancing spatial cognition, the combination offers new ideas and new interface to spatial information visualization representation.Secondly, it discusses basic characteristics, application and structure of outdoor AR spatial information representation. In order to provide a reference to build and improve research framework, it analyzes existing technologic prototypes from application-driven perspective and puts forward a research framework and discusses four key technologies:display device, tracking and positioning, virtual-real combination and HCI.Thirdly, it proposes and practices a hybrid registration method based on GPS and 3D electronic compass. First of all, it introduces definition and conversion of world coordinates, camera coordinates, virtual coordinates, image plane coordinates and pixel coordinates and advances calibration method to get camera external and internal parameter. Secondly, it improves existing fiducial detection registration method in indoor AR by replacing relative scene coordinates with world coordinates to augment real scene through spatial information and puts forward a registration method based on visual fiducial detection. Lastly, it proposes a hybrid hardware registration method which uses GPS and 3D electronic compass to obtain position and pose of user’s head. Given position and pose in world coordinates of virtual objects through HCI, it obtains position and pose of user’s head to get transformation matrix from world coordinates to camera coordinates. Known internal camera parameters, transforming camera coordinates to 2D image plane coordinates, it achieves combination of virtual objects and real scene image.Fourthly, it proposes and verifies registration methods using affine representation. In order to avoid complex and tedious process of position tracking and camera calibration, According to observation from Koenderink and van Doom, Ullman and Basri in 1991 which is given a set of four or more non-coplanar 3D points, the projection of all points in the set can be computed as a linear combination of the projection of just four of the points. The paper sets up global affine coordinate system based on world coordinates, camera coordinates and virtual coordinates. It extracts four feature points from scene image and calculates global affine coordinates of key points of virtual objects. Then according to a linear homogeneous coordinates of four feature point’s projection, it calculates projection coordinates of key points of virtual objects. In addition, it proposes an approach to get pixel relative depth to hidden surface removal. In the end, it analyzes dynamic registration error and the ways to reduce.Finally, virtual-real combination and HCI which is discussed from three aspects of geographical objects modeling, virtual-real combination and HCI, are also important characteristics of outdoor AR spatial information representation which reflects demands of display quality and natural, efficient way of "User-centric" HCI更多还原