【作者】 胡寅；

【导师】 李德华；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2005， 博士

【摘要】 随着计算机技术的发展,工业生产、影视制作、服装设计、医疗诊断和军事训练等领域提出了一个越来越迫切的要求:快速获取物体的三维模型。三维扫描仪就是一种能够直接得到物体的原始三维信息的计算机输入设备。这种从实物样件获取产品数学模型的技术,称之为逆向工程。我们研制的三维扫描仪系列产品集光学、机械、电子、自动控制、数据采集、图像处理等诸多领域的知识为一体,是具有重大应用价值的高科技产品。本文在国家自然科学基金(69775022)和“863”高技术研究发展计划(863-306-ZT04-06-3)的资助下,对三维扫描仪系统进行了理论研究和技术实现,重点研究了三维信息获取和逆向工程中的几个关键技术:系统定标、数据修补、点云压缩、三维重建、网格简化等。本文首先系统总结了现有三维信息获取技术,简要介绍了常见的几种三维扫描仪产品和今后的发展趋势,提出了三维数字化基本流程和关键技术。三维扫描仪按照信息获取方式的不同可分为接触式和非接触式两大类。接触式使用测头直接触碰物体表面,根据测量装置的空间几何结构得到测头的坐标; 非接触式主要基于计算机视觉原理,从摄像机拍摄的图像中获取目标的三维信息。考虑到不同的应用场合,我们分别研制了接触式多关节机械臂三维扫描仪(3DLCS-400)和非接触式结构光彩色三维扫描仪(3DLCS-200)。本文简要介绍了这两种三维扫描仪的系统原理、系统构成以及工作流程。系统结构参数偏差会导致坐标测量误差,为提高测量精度,对三维扫描仪进行定标,以获得尽可能准确的实际结构参数是非常必要的。本文针对非线性系统的多参数定标问题,结合最小二乘法和遗传算法各自优点,提出了一种收敛速度快,又有较高精度的混合智能算法。首先通过改进的最小二乘法计算得到问题的次优解,以此作为遗传算法的基因中心值,并将基因范围动态缩小进行进化计算,从而获得最优解。该算法解决了传统方法遇到奇异阵无法求逆的困难,克服了一般遗传算法效率太低的缺点,大大提高了定标的速度和精度。在三维扫描仪对物体扫描过程中,由于各种因素的限制,可能导致物体某些更多还原

【Abstract】 Along with the development of computer technology, an urgent request, that is how to obtain 3D model of objects, is put forward in industrial production, movie making, apparel design, medical diagnosis and military training etc. 3D scanner is a kind of input equipment which can acquire the 3D data of an object directly. This technology comes to be known as Reverse Engineering. 3D scanner realtes to such subjects as optics, mechanics, electronics, automatic control, image process etc and has much significance on many applications. In this dissertation, some key technology used in 3D scanner are developed including 3D acquiring, system calibration, data repairing, point cloud compression, 3D reconstruction and triangle mesh simplification. My work has been supported by National Natural Science Fundation (69775022) and National 863 Project (863-306-ZT04-06-3). First, to the fields of 3D acquiring, it systematically concludes the popular 3D acquiring methods and some products of 3D scanner. The basic process and key technology are also presented. 3D scanner can be divided into two types according to its acquiring method: contact and non-contact. The former uses probe to touch the surface of objects and obtain the coordination from geometry structure. The latter gets 3D information from a series of photographs of video camera based on computer vision theory. Considering different applications, we developped contact multi-joint 3D scanner(3dlcs-400) and non-contact constructed lighting 3D scanner(3dlcs-200). The theory, constitute and work process are introduced. System parameter errors will result in the coordination errors. In order to improve measuring accuracy, it is necessary to calibrate the structure parameters. Combined with fast convergence and high accuracy, a hybrid intelligent algorithm is proprosed to solve the parameter calibration of the nonlinear system. The results that computed through the improved least-square algorithm become the the original values of the genetic algorithm where gene range could be dynamically changed. The experiment shows that the hybrid algorithm is valid in practical application. Because of some restrictions, incomplete data is possible to appear in the scanning process. Data repairing based on genetic algorithm and BP neural network is investigated to solve the problem. With the improvement of accuracy, the number of point cloud increases quickly and needs to be compressed. The simplest method uses sampling at the cost of quality, so it is not fit for complex object’s surface. In this dissertation, an adaptive 3D Splitting method for compressing of the data on the spacial contours is presented. Triangle mesh model is selected to build boundary representation. The reconstruction methods with orderliness data and scattered data are discussed. The new way uses a set of multilayer parallel planes to bed the set of scattered points. First, we regulate the scattered points into a serial of contour lines. Seconed, we sort the points in the same of contour lines in anticlockwise. Final, we reconstruct the object with the way of parallel contour lines. Nowadays, the quantity of triangle mesh becomes more and more to meet the demand of reality sensation. But complex mesh model is adverse to storage, transfer and render. Progressive mesh is a good method to simplify meshes. In PM form, an arbitrary mesh is stored as a much coarser mesh together with a sequence of detail records that indicate how to incrementally refine exactly back into the original mesh. The PM representation naturally supports progressive transmission, offers a concise encoding of itself, and permits selective refinement. On the basis of PM, a new method which collaspe half edge and keep the color attributes is given. It not only avoids the complexity of computing new vertex, but also embodys the effect of color. The practical results appear perfect.更多还原