【作者】 周漾；

【导师】 郑顺义；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2013， 博士

【摘要】 文物是人类社会活动中遗留下来的具有历史、艺术、科学价值的遗物和遗迹。馆藏文物纹理重建与组织关键技术研究对于文物的传承、鉴赏和保护管理具有重要的理论意义和应用价值目标的三维重建包括几何重建和纹理重建两部分,本文针对纹理重建,假设已有馆藏文物对象的结构光扫描点云、扫描同步影像和单反相机拍摄的高分辨率影像,目的是基于这些数据完成馆藏文物的高质量、高保真的纹理重建、组织与展示工作。其中,本文完成的研究工作包括：1)结构光系统扫描点云的三维构网。在众多几何重建手段中,基于结构光扫描系统或激光扫描系统是获取重建对象表面密集点云的最稳健的方法之一。然而几何重建的最终目标是物体的几何模型,对于扫描获取的点云数据,还需要进行离散点表面重建。本文总结分析了现有的三维构网方法,选择了其中七种算法进行了详细的对比试验,最终选择了泊松构网方法对扫描点云进行表面重建。2)基于结构光系统扫描同步影像的高分辨率影像的配准。结构光方法本质上仍是一种基于影像的测量方法,所以其测量的过程中能同步采集扫描对象的影像。经过点云拼接,这些影像的外方位元素只要相应的进行相似变换,就能得到其相对于最终几何模型的外方位元素。利用这些已知方位元素的同步影像完成高分辨影像的自动配准是本文选择结构光系统进行重建的最重要原因。本文以扫描同步影像为参考基准,使用匹配加光线投射的方法在高分辨率影像与几何模型间自动寻找同名点,最后采用稳健后方交会的方法完成配准,整个过程几乎不需要人干预,精度和可靠性高。3)多张影像的无缝纹理映射。当完成了多张高分辨率影像与几何模型的配准,由于影像间不可避免的色彩、光照等差异,在最终的纹理图集上总是存在纹理接缝。本文回顾总结了现有的无缝纹理处理方法,基于泊松影像编辑和色彩变换技术,提出了一种新的以面为处理单元的无缝纹理辐射处理方法。理论分析和试验证明了相比标准的像素级泊松影像编辑方法,本文方法更适合于三维模型的纹理重建。4)纹理组织与纹理模型的简化。随着硬件设备的发展,扫描点云越来越密集,数码相机的分辨率也逐年提高。因此高效率的组织纹理数据和高质量的简化纹理模型对数据传输、保存和展示都意义重大。在纹理组织上,本文在现有的贪婪法纹理打包基础上,增加了纹理块矩形融合,进一步提高了效率和稳健性。在模型简化问题上,针对纹理模型的简化后的纹理恢复问题,本文首次提出了一条完整而高质量的纹理重采样解决方案,不但丰富了纹理模型的简化方法,在和现有方法的对比上,本文方法也表现出了不同的特点和优势。5)纹理重建应用。本文开发了一套结合结构光系统的纹理重建软件,进行了大量的馆藏文物、工艺品等纹理重建工作。而随着移动设备和互联网的发展,当前三维模型的展示已经不限于PC设备。本文基于OGRE开源三维引擎,开发了一套iPad三维模型浏览软件用于移动设备的展示。最后,本文基于重建的纹理模型,提出了一种新的真正射影像采集方法,该方法采集的真正射影像,不仅无需镶嵌,而且是无缝的。更多还原

【Abstract】 Cultural Relics are the heritages with historical, art and scientific values. The3D digital reconstruction, which is an important method to record, preserve, inherit and appreciate the cultural relics, is now becoming more significant and challenging.The3D reconstruction for targets includes two steps of geometric and texture reconstruction. Aiming at the texture reconstruction, this paper proposed a complete solution for the cultural relics in museums. The input data of our method is the point clouds with synchronous images from structure-light scanning system and the high resolution images from SLR cameras. Our method has the merits of high quality, high fidelity and high efficiency, in which the major contributions include:1.3D surface reconstruction from scanning point clouds. Among the general-used methods in3D measurement, structure-light scanning system is one of the most stable ways to gain the dense and uniform points from objects’surface. But as the final result of geometric reconstruction should be the mesh model, it needs to construct the triangular mesh from the point clouds. After summarizing the existing surface reconstruction methods, we chose7approaches to carry through the real data experiments. And finally, due to the superiority in handling noise and non-uniform samplings, Poisson surface reconstruction method is chosen to do geometric reconstruction for the point clouds from structure-light system.2. The automatic registration between high resolution images to geometric model. Since the structure-light system is still an image-based measurement method, it can capture a synchronous image during one scan. These synchronous images can be registered easily to the final geometric model along with the point clouds alignment. First, we regard these synchronous images with known parameters as the reference. Then the corresponding point pairs between high resolution images and geometric model are calculated through image matching and ray casting techniques. At last, the exterior parameters of high resolution images are accurately estimated through robust resection. The whole process is automatic and efficient, and experiments proved our method.3. Seamless texture mapping from multi-view images. After registering sufficient high resolution images, seamless texture mapping is another challenge, due to the inevitable differences between images in color, illuminance etc. First, we reviewed the existing methods in removing texture seams. Then we proposed a new photometric processing method of face-wise Poisson blending for removing photometric discrepancy on seam-lines, which is based on standard Poisson image editing approach combined with color transfer techniques. Compared with the existing methods, theoretical analysis and experiments proved that our method is more appropriate for3D seamless texturing.4. Optimizing texture maps and simplifying the texture model. As the development of hardware, the density of scanning points and the resolution of images become increasingly higher. For reducing the redundancy of texture maps, we reformed the existing texture packing methods by texture block merging. And then, we proposed a complete and high quality solution to resampling the texture for the simplified geometric model from the original texture maps. Our method not only enriches the simplification methods for texture models, but also shows advantages in comparison to the existing methods.5. The application of texture reconstruction. We developed a set of software for texture reconstruction, and do a lot of3D reconstruction for cultural relics in museums, art and crafts etc. And for the exhibition on mobile platforms, we developed an iPad App based on the OGRE3D graphics engine. At last, we proposed a true-ortho photo generation method based on the result of texture reconstruction.更多还原