【作者】 闫龙；

【导师】 赵正旭；

【作者基本信息】 山东大学 ， 机械电子工程， 2008， 博士

【摘要】 本文设计研制了基于摄影测量原理的便携型立体视觉测量系统,该系统在不借助激光、光栅、反差增强剂等辅助设施的情况下,利用普通图像和立体视觉测量的方法获取所测物体的形貌及三维深度信息,能够满足快速测量及建模的大部分需求,具有体积小、成本低、开发容易、携带方便等优点。对便携型立体视觉测量系统的相关技术进行了研究,包括摄影测量器材的选择、系统结构组合等硬件平台的搭建;图像获取与处理、立体匹配、三维深度信息提取、深度数据存储等软件模块的实现;系统标定及测量精度的分析和评价等,最后通过实例示范了系统的使用方法及三维建模的过程,系统可准确完整地获取小型物体(包括纹理单一的机械零件)的三维形貌信息,测量精度能够满足逆向工程、虚拟工程、动画制作等领域中测量及建模的要求。课题的主要工作和创新之处包括以下几个方面:1、通过引入数字量化效应修正了立体视觉系统测量模型,估计了系统的有效视场,分析了系统的测量精度,并对系统的结构提出了优化方案。在工程中可以根据摄像机参数、基线长度、测量距离等参数估计测量误差;也可根据测量精度要求,指导立体视觉测量系统的结构设计。2、提出了针对单一纹理表面的立体匹配方法,扩大了摄影测量的适用对象和应用范围。通过边缘检测、极线几何和立体图像边缘的相关性确定图像的匹配约束和搜索范围,结合不同的匹配窗口和对局部区域进行由粗到精的渐次匹配、并利用全局松弛和滤波得到高密度高质量的视差图。实验证明该方法提高了匹配效率,一定程度上克服了表面反射和单一纹理的困扰,提高了测量数据的质量,所得结果可用于机械零件的三维重建。3、提出了针对摄影测量所得点云数据的滤波和精简的新方法。摄影测量所得点云数据中噪声较多,脉冲噪声和随机噪声同时存在,孤点与成段的数据段难以区别特征点与噪声,普通方法不能很好地区分。针对该问题提出了一种基于形态学的滤波算法。实验证明该方法运算速度快,不仅能滤除孤点、成段的噪点,还能对某些遮挡或陡峭区域进行一定程度的修补。摄影测量所得数据量较大,一次拍摄所得的原始数据通常有几兆到几十兆字节,这种情况无论是对于数据量的传输还是后期的表面重构都带来很大的困难。为了降低数据的冗余度,本文采取图像处理的手段对点云数据进行精简,使用阈值分割的方法自动确定精简比,实验证明精简后的数据保持了原有特征,可用于深度信息的保存及后续步骤的进行。4、改进了基于二维图像的三维重构过程,提出了一种三维图像格式,利用普通二维图像结合摄影测量所得数据生成三维图像,既拥有二维图像供查看和传播的便利性,又拥有三维深度信息的完整性,可供三维重构使用或用于二次开发,为三维图像提供了一种新思路,拓宽了视觉测量系统的应用领域。5、开发了一套新的便携型视觉测量及建模系统,在几个关键技术上提出了解决方案,基本实现了小型物体三维形貌的获取功能。以机械零件为例对系统及各种算法进行研究,证明了设备、算法和系统用于测量和建模的可行性。该系统低廉的价格、灵活的开发方式使其具有较大的使用群体和更广阔的应用范围。可用于场景、中小型物体特别是机械零件的三维测量和建模,在逆向工程设计、多媒体动画创作、科研分析以及文物玩具等模型的三维数字化方面具有广阔的应用前景。更多还原

【Abstract】 This thesis documents the research into the design and development of an effective portable vision system for measuring and modeling. The system achieves depth information by stereo vision system without laser or raster. It meets most of the needs of rapid measuring and modeling in engineering entertainment and business. Portable stereo measuring system based photogrammetry is small in capacity, cheap and easy-developed.It also documents the research into related techniques which include device composition, system configuration and assembly, system accuracy analysis, calibration, image acquisition, image processing, stereo match, depth data extraction and data storage. Examples are provided to verify the correctness and feasibility of the system and key techniques that are created. The accuracy can meet the need of measuring and 3D reconstruction area.In summary, the research has made following contribution:(1)Modifies the accuracy estimation models by adding in digital quantization effect, computes the effective visual workspace, analyzes system measuring accuracy, and optimizes the system structure. With these works, measuring errors can be estimated with parameters of camera, length of baseline and measuring range in the engineering. The system can also be redesigned with these models.(2)Promotes stereo matching algorithm for monotone texture, expands application range of vision system in some extent. This thesis sets the matching and searching constraints by edge detecting, epipolar geometry and correlation of stereo pairs, then achieves the matching result by iterative area match from coarse to fine. Experiments have verified that this algorithm can not only enhance matching accuracy, the results are more accurate and reliable and can be employed for 3D reconstruction in reverse engineering.(3)Proposes filtering and compression algorithm for cloud data achieved in photogrammetry. Pulse noise and random noise present simultaneously, while single point is difficult to tell from features and noises. The thesis presents a method of noise filtering for data cloud based on morphology. Morphology based filtering is verified effective for the data cloud processing in photogrammetry. It not only filters noises but also repair some flaws. The depth data is very large. One piece of image contains tens of million of bytes. It is difficult for information transformation or surface reconstruction. The thesis reduces cloud data based on image processing methods. Experimental test shows that the data can be reduced to 60% without losing vital information, especially the most important features. This has been used to conceive and elaborate the following up research for example, surface reconstructions. Cloud data compression and reduction based image processing technologies is creative in this area. Template computing, histogram, threshold segmentation and uniform grid are composed together.(4)Improves 3D reconstruction flows based 2D images. This research composes image and depth information into one image file format, which is called 3D image. Reduction, compression and recoding make the image pair more slender and depth data can be saved in image for ever, which implements images as data blocks. The 3d image not only can be used as common image for display, but also can be used for facilitating 3d reconstruction. It has features of both raster images and vector graphics, preceded to the transforming of two objects and wide application for image based on modeling and reconstruction.(5)One portable vision system is developed for 3D measuring and reconstruction of mechanical parts. Several key techniques are overcome in new solutions. Examples prove that the algorithms and theories are right and suitable for this system.This system will have more users and more application for its low price and smart developping styles. It can be used in 3D measuring and reconstruction of scene and mechanical parts. Examples are provided as tests to verify the theory and application of the system, which proves that this system can be used in virtual engineering, reverse engineering and modeling of rapid prototyping.更多还原