3D reconstruction of an outdoor archaeological site through a multi-view stereo technique
3D reconstruction of an outdoor archaeological site through a multi-view stereo technique
复制标题
通过多视图立体技术对室外考古遗址进行 3D 重建
DOI：
--
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
Antonio Lamarca
中科院分区：
文献类型：
--
作者：
M. Muzzupappa;A. Gallo;F. Spadafora;Felix Manfredi;F. Bruno;Antonio Lamarca
本文描述了在一个特别复杂的测试案例上对常见多视图立体技术的实验：土耳其 Kyme Eolica 考古遗址的“Via Colonnata”。该研究表明，无需使用激光扫描仪、无人机或氦气球等任何专用设备，只需使用数码相机和开源软件即可创建数十平方米区域的详细 3D 模型。本研究中实现的重建过程解决了与大面积重建以及随后在几何模型上映射纹理相关的一些最相关的问题。特别是，我们为采集阶段提出了一些指导原则，有助于减少与 3D 几何创建和纹理映射相关的后续问题。在预处理阶段，我们提出了一种基于对每个图像对相关视差图的分析来过滤不重要区域的自动化技术（距离当前视点较远的区域被屏蔽，以获得没有伪影/缺陷的3D模型）。对于纹理映射过程，为了减少重叠区域中的平均和混合操作造成的模糊，我们提出了一种自动识别要投影到 3D 模型上的最合适的图像子集的方法。
This paper describes the experimentation of a common multi-view stereo technique on a particularly complex test case: the “Via Colonnata" in the archaeological site of Kyme Eolica in Turkey. The study demonstrates that it is possible to create a detailed 3D model of an area sized tens of square meters without the need to use any dedicated device like laser scanners, drones or helium balloons, but just employing a digital camera and open source software. The reconstruction process implemented in this study addresses and solves some of the most relevant problems related to the reconstruction of large areas and the subsequent mapping of a texture on the geometrical model. In particular, we suggest some guidelines for the acquisition phase that help to reduce the subsequent problems related both to 3D geometry creation and texture mapping. In the pre-processing phase, we propose an automated technique for filtering of unimportant areas, based on the analysis of the disparity maps related to each image pair (the farther areas from the current point of view are masked out, in order to obtain a 3D model free of artifacts/defects). For the texture mapping process, in order to reduce the blur resulting from averaging and blending operations in overlapping areas, we propose a method that automatically identifies the most appropriate subset of images to be projected on the 3D model.