QQ客服 官方微博 反馈留言

题名	基于二维达曼光栅的物体三维轮廓测量研究
作者	韦盛斌
学位类别	博士
答辩日期	2015
授予单位	中国科学院上海光学精密机械研究所
导师	周常河
关键词	三维轮廓测量术 达曼光栅 双目匹配 点云配准
其他题名	Three-dimensional Profilometry using Two-dimensional Dammann Gratings
中文摘要	三维轮廓是描述物体特征的最重要信息之一。作为获得物体三维轮廓信息的重要手段，光学三维轮廓测量术在过去十年内被广泛地运用于机器人、制造业、生物医疗、航空航天以及考古等传统应用领域中。伴随着影视娱乐、三维打印、人机交互、虚拟现实和可穿戴设备等新兴应用领域的的发展，光学三维轮廓测量术正面临着新的挑战：低成本、体积小、便携，同时还要具备一定的测量精度。 虽然利用数字投影结构光的三维轮廓测量术一直在发展，但是现阶段难以满足所有测量要求。因此需要发展利用衍射光学元件产生结构光的三维轮廓测量术。本文在总结了现有的利用数字投影和利用衍射光学元件投影的三维轮廓测量术后，提出了一种基于二维达曼光栅（Dammann grating）的三维轮廓测量术，该测量术具有可手持测量、体积小、信噪比高和成本低等优点。 本博士论文通过从整体到细节的方式介绍基于二维达曼光栅的三维轮廓测量术，主要从几个方面介绍本论文的研究工作： 1）首先，从整体介绍了基于二维达曼光栅的三维轮廓测量术的测量原理。在测量中，由达曼光栅产生的激光点阵投影到物体表面，双目相机采集物体表面的激光点阵，然后所得的图像经过激光光斑提取、双目匹配、点云计算和点云处理等数据处理主流程，即可得到物体的三维模型。在样机搭建阶段，我们分析样机参数对于测量指标的影响，设计出了两台参数和测量指标不相同的样机，其中一台用于低成本化可行性验证，另一台用于高精度测量。 2）其次，我们从细节介绍基于二维达曼光栅的三维轮廓测量术的核心算法——双目匹配算法。在基于图像灰度信息的双目匹配方法失效的情况下，我们提出了一种结合极线约束、视差约束和射线约束这三种几何约束的双目匹配算法。其中，由于达曼激光点阵的引入，我们首次提出了射线约束。在具体研究过程中，我们通过分析算法中的关键参数的方式，反复改进算法以提高性能。我们分别提出了一种改进的双目匹配算法和一种快速的双目匹配算法。 3）最后，我们从细节介绍了基于二维达曼光栅的三维轮廓测量术遇到的另一个困难——稀疏点云的配准问题。针对稀疏点云，我们提出了一种基于点云单应性的、采用“点-面”距离的迭代最近点算法（BC-p2s-ICP）。理论方面，我们首先介绍了单应性“点-面”距离的计算方法，然后提出了利用KD树进行算法优化，最后给出了坐标变换矩阵的计算方法。在实验方面，我们将BC-p2s-ICP算法和三种经典ICP算法进行比较，实验结果表明BC-p2s-ICP算法具备较强的鲁棒性、较快的收敛速度和较高的收敛精度。 需要强调的是，以上的研究内容是围绕着三维重建这一主线，按照从整体到细节的方式进行讨论的，每一项内容并不是孤立的。本博士论文的讨论内容包括三维重建的算法理论和大量实验验证，对后续研究和技术产品化具有重要的指导意义。
英文摘要	Three-dimensional (3D) profile is one of the most important descriptions of objects. As an important method to acquire 3D profile of an object, optical 3D profilometry has been applied in traditional areas such as robotics, manufacturing, biomedical, aerospace, and archaeology during the last decade. With the developments of new technologies such as studio entertainment, 3D printing, human-computer interaction, virtual reality and wearable devices, optical 3D profilometry faces new challenges: low cost, small volume, convenience and relative high measurement accuracy. It is not easy for those 3D profilometries using digital projection structured lights to satisfy all the measurement requirements. Thus it’s necessary to improve 3D profilometry which uses diffractive optical elements (DOE) to generate structured light. In consideration of the existing 3D profilometries whose structured lights are generated by digital projection or DOE, we propose a new 3D profilometry based on a two-dimensional (2D) Dammann grating, which has advantages of portable measurement, small size, high signal-noise-rate and low cost. This doctoral dissertation introduces the new 3D profilometry based on a 2D Dammann grating, which includes the following aspects: 1) First, we introduce the basic measurement principle of this new 3D profilometry in general. During measurements, laser dot array, which is generated by a Dammann grating, is projected onto the surface of objects, and then it is captured by a binocular camera. The processes from original images to 3D models mainly include laser spots extraction, binocular matching, point clouds computing and point clouds processes. At the stage of prototypes building, we analyze the influence of parameters to measurements, and we designed two prototypes with different parameters and measurement requirements. One prototype is set up to verify the feasibility of low cost, and the other one is set up for high accuracy measurements. 2) Second, we introduce the key algorithm of 3D reconstruction — binocular matching. While those existing matching algorithms based on grayscales are invalid, we propose a new algorithm using three geometrical constraints: epipolar constraint, disparity constraint and ray constraint. Among these three constraints, ray constraint, which is a new constraint brought in by the Dammann grating, is proposed for the first time. In the researches, we have been keeping improve our algorithm for better performance by analyzing the key parameters in binocular matching. An improved algorithm and a fast algorithm are proposed individually. 3) Third, we discuss another problem that our scheme faces — the registration of sparse point clouds. For sparse point clouds, we proposed a new iterative closest point (ICP) algorithm based on the biunique correspondence of point clouds and ‘point to surface’ distances (BC-p2s-ICP). We introduce the computing of biunique correspondent ‘point to surface’ distances, and then we propose an optimization using KD tree, and finally we present the formula of coordinate transformation matrix. In experiment, we compare our BC-p2s-ICP to three represented ICP variants, and the experimental results demonstrate that BC-p2s-ICP has stronger robustness, higher convergent speed and higher convergent accuracy. It is very important to emphasize that the aspects above are discussed around the issue of 3D reconstruction, so each aspect is not isolate. The doctoral dissertation includes 3D reconstruction algorithms and experimental verification, which is important for further studies and applications.
语种	中文
内容类型	学位论文
源URL	[http://ir.siom.ac.cn/handle/181231/15905]
专题	上海光学精密机械研究所_学位论文
推荐引用方式 GB/T 7714	韦盛斌. 基于二维达曼光栅的物体三维轮廓测量研究[D]. 中国科学院上海光学精密机械研究所. 2015.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。