| 题名 | 结构光双目三维测量技术研究 |
| 作者 | 刘昆 |
| 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 | 三维测量 结构光 光点阵列 立体匹配 点云配准 |
| 其他题名 | Research on Binocular Three-dimensional Measurement Technology Using Structured Light |
| 中文摘要 | 近年来，非接触无损三维测量吸引了越来越多的关注并在众多领域得到了广泛地运用。非接触式光学测量物体表面的方法现在被广泛运用于各个行业，包括：医疗卫生保健、珍贵文物数字化、逆向工程、工业检测、计算机图形学与特效、智能机器人导航、娱乐行业等方面。非接触三维测量技术是一种基于光学手段和图形图像处理分析方法的非接触式测量，其运用计算机图形图像学的理论来数字化再现物体的三维形貌，是现在计算机视觉研究的主流方向之一。非接触三维测量技术一般可分为主动与被动两类。主动式技术是指将额外的能量投射至物体，借由能量的反射来计算三维空间信息。目前常用的非接触主动光测量方法有：飞行时间法、三角测距法、结构光法等。被动式扫描仪本身并不发射任何辐射线，而是以测量由待测物表面反射环境光的方法，达到预期的效果。被动式方法受环境照明与目标物体自身表面属性影响很大。 本文首先介绍了三维测量技术的发展现状。随后讨论了几种常用三维测量技术，对其原理，使用与优缺点做了分析。结构光法通过向待测物体投射一系列经过特殊设计的图案来实现三维重建。这些经过特殊设计的图案非常有助于提升包括测量效率、测量精度在内的系统指标。双目立体视觉是机器视觉的一种重要形式。它是基于视差原理并利用成像设备从不同的位置获取被测物体的两幅图像，可减少单目观测存在的观察死角与遮挡现象。鉴于结构光三维测量技术的诸多优势，本博士论文所讨论的三维测量装置均采用双目结构光方案，并主要开展了以下几个方面的研究工作： 1. 搭建了一种基于结构光投影的双目三维形貌测量装置。该装置主要由两个摄像机与一台置于两个摄像机中间的光学引擎组成。测量时，光学引擎向待测物体投射一系列二值带通伪随机图案。双目摄像机同步捕捉待测物体被这些图案照明的图像，得到一系列立体图像对。随后运用时域立体匹配算法提取出同源点，进而获得同源点的视差。一个经过标定的双目系统，可以根据视差利用三角原理重建物体的三维形貌。相对采用光点阵列的扫描方式，该方法可一次性获取足够密集的数据点，减少点云匹配的计算次数。 2. 针对密集立体匹配过程纷繁耗时的问题，本文提出了两种搜索区间优化策略。一种对视差进行预估，将同源点搜索范围限制在预估位置周围。另一种借由一幅内嵌识别块的条纹图案，将待测物体分割为多个条纹，在判定了条纹的对应关系后将同源点搜索范围限制在由条纹宽度限定的一个相对小的区域内。实验中，采用视差进行预估的搜索优化策略减少了一半的运算量，而借由条纹图案的搜索优化策略将运算量缩小至原来的25%。 3. 搭建了一种基于二维Dammann光栅的点阵结构光扫描方式。该装置采用小型半导体激光器与二维Dammann光栅配合产生二维激光点阵。Dammann光栅将光源的光强等分成多份，两个摄像机同步采集二维激光点阵投射到待测物体表面形成的图案。提出了一种基于衍射级次分析的光点匹配方法对所获取的激光点阵进行立体匹配，通过三角原理获得三维点云。我们对物体多次扫描，最后将多幅点云拼接形成密集点云。该方法对环境要求不严格，一般室内环境下小型半导体激光器产生点阵足够明亮，不会被环境光湮没。 运用上述方法，我们测量了包括石膏像、瓷器、粗糙表面的透镜毛坯等物体，获得了待测物体具有较高精度的密集点云，验证了其适应性、可靠性与实用性。 |
| 英文摘要 | Non-contact and non-destructive three-dimensional (3D) digitalization of object has attracted more and more attention and applications over the past decade. Several examples of such applications are: computer aided medical procedures, 3D documentation of precious cultural relics, reverse engineering of existing mechanical device, inspection and control of shape quality in industrial production system, computer graphics and special effects, intelligent robot navigation, movies and video games in entertainment industry and so on. Non-contact 3D measurement technology is based on optical method and digital image processing method. The technology uses computer graphics theory to reconstruct digital 3D model of objects, and it becomes one of the main directions of computer vision research now. A number of technologies have been developed for digitally acquiring the shape of 3D objects. A well established classification divides them into two types: contact and non-contact. Non-contact solutions can be further divided into two main categories, active and passive. Active scanners emit some kind of radiation or light and detect its reflection or radiation passing through object in order to probe an object or environment. Time of flight method, triangulation method, and the structured light method are commonly used active non-contact optical measurement methods. Passive 3D imaging solutions do not emit any kind of radiation themselves. They rely on detecting reflected ambient radiation. The performance of passive scanners is greatly impacted by ambient lighting and the surface scattering properties of object surface. When facing bleak environment or some featureless objects, passive methods could easily fail. In this paper, we introduces the some commonly used 3D measurement technology such as time of flight method, triangulation method, structured light method and binocular vision and so on. The principles, operation and the advantages and disadvantages are analyzed. Structured light method is an important class of 3D surface imaging techniques that use active light source to project 2D pattern(s) onto the surface of object in the scene. The projected image pattern(s) is specially designed in order to maximize the efficiency, accuracy, and some other system performance indexes of a 3D surface imaging system. Binocular stereo vision is the extraction of 3D information from digital images. By comparing information about a scene from two vantage points, 3D information can be extracted by examination of the relative positions of objects in the two images. Active stereo vision is a form of stereo vision which actively employs a light such as a laser or a structured light to simplify the stereo matching problem. This dissertation adopts structured light binocular program for its advantages. Mainly research works are listed as follows. 1. We develop an active binocular 3D scanning setup using band-pass random patterns projection. A series of N binary band limited patterns will be projected to the target object. Two cameras capture the image of the target object synchronously. The band-pass white random patterns are specifically designed to reduce the effects of indirect illumination, and it is robust to standard challenges in scanning systems such as defocus and low camera-projector pixel ratio. While this approach uses structured light patterns that increase the number of required projected images, similarity measures identify homologous points purely in the temporal domain. Finally, an accurate point cloud of optical element can be obtained by triangulation methods and subpixel interpolation. Experimental results verify the robustness, easy operation and high reconstruction accuracy of the setup. 2. Two optimized stereo matching methods are proposed. Traditional dense stereo matching algorithm is time-consuming due to long search range and high complexity of similarity evaluation. We project a binary fringe pattern (BFP) in combination with a series of N binary band limited patterns (BBLP). In order to prune the search range. An initial matching will be executed before exhaustive matching and evaluate similarity measure using logical comparison instead of complicated floating-point operation. As a comparison, we introduce another search range optimization method which is based on an estimate of disparity. The simplified algorithm based on estimated disparity reduces the number of similarity measure calculations by half. By contrast, the search strategy based on initial matching has more excellent performance that three-quarters of the calculations will be excused. Experimental results verify the computational efficiency and matching accuracy of these methods. 3. We develop another binocular three-dimensional measurement system using Dammann grating. A laser diode and a Dammann grating are employed to generate a regular and square laser spot array. Dammann array illuminator is placed between two cameras. During the measurement, a 64×64 laser spot arrays will be projected toward the target object and captured by two cameras simultaneously. Stereo matching which is based on the diffracted orders analysis of the laser spots will be performed to search the homologous spot. Stereo matching is easily performed based on a fact that laser spots with the same diffraction order are homologous ones. One frame three-dimensional point cloud can be obtained using the disparity of homologous points by triangulation methods. Finally, 3D point clouds belong to different frame which represent different view of the object will be registered to build up an integral three-dimensional object. This method is insensitive to ambient light. Using the above method, we measured some objects such as plaster, porcelain, a semi-finished optical product with rough surface. Dense point clouds with high accuracy are obtained, and shows suitability, reliability and practicability of structured light method. |
| 语种 | 中文 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.siom.ac.cn/handle/181231/15904]  |
| 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 | 刘昆. 结构光双目三维测量技术研究[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。
修改评论