多目视觉坐标测量的关键技术研究

【作者】 杜娜；

【导师】 方漪；

【作者基本信息】 青岛大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 立体视觉坐标测量技术作为一种以立体视觉和光学成象原理为基础的非接触式三维测量技术,以其非接触性、并发测量速度快以及精度高等优点广泛应用于工业三维测量、虚拟现实建模、反求工程等不同领域。 本论文针对立体视觉坐标测量的关键技术,包括立体匹配和数据拼接,进行了研究,主要研究内容包括以下几个方面: 1.多目视觉坐标测量的原理和测量模型。采用多目视觉模型进行三维测量,针对摄象机线性模型,利用带有网格线的标定物线性标定摄象机。 2.有效测量域的数学模型和求解。可测量的点即可匹配的点必定位于系统的有效测量视场内。本文对有效测量域进行建模,同时给出其参数的求解方法,并将测量域模型应用于匹配算法剔除误匹配。 3.结合支持度判断的匹配方法。匹配是视觉坐标测量的关键,是要在多幅图象中确定哪些象点对应场景中同一个点。匹配算法影响测量系统的准确性与稳定性。对匹配算法的基本要求有两个,算法的精度或可靠性以及算法的效率。针对这些要求,本文提出了两种方法,一种是通过对有效测量域的求解,减少匹配遍历的空间,提高算法效率,并且能够检测到测量域之外的误匹配,在一定程度上也能提高测量的精度。另一种是针对空间匹配的几何模型,结合多种几何与逻辑约束,检测并排除误匹配,提高测量结果的准确性和测量系统的稳定性。本文给出的实验结果说明了算法的有效性,并且算法时间复杂度为线性函数。 4.不同视点三维测量数据的拼接融合。视觉坐标测量系统在对大视场范围的场景或表面复杂的物体进行测量时,单次测量常常不能完整准确的采集整个表面。为了保证测量范围和测量精度,将测量空间划分为子空间进行多次测量,然后根据公共点对各次测量数据进行拼接配准,得到统一坐标系下的整体测量数据。本文在对拼接方法进行研究的基础上,对于不少于三个不共线公共点,提出通过构造虚拟点的方法线性地求解配准变换矩阵。 其中,后三部分是本文的研究重点。更多还原

【Abstract】 Vision Coordinate Measurement technique is a non-touch measuring technique based on stereo vision and optics. It is widely used in 3D measurement in industrial, virtual reality modeling, reverse engineering and other fields for its high speed and high accuracy.Stereo matching and data registering are two techniques of great importance in Vision Coordinate Measurement. And this thesis puts emphasis on the two topics. The researches developed in this paper include four parts.1. Systematic model for measuring and Camera calibration. Multi-view stereo vision model is considered as the computing model of the vision measuring system. And linear cameras model is taken to calibrate system parameters.2. Constructing mathematic model of valid measurement space, analysis and calculation on it All the image points which can be matched with another one must lie in the valid measurement space. This paper models the 3D space with spatial geometry and gives the solution. This paper also uses the model to detect and get rid of mismatches.3. The matching algorithm with support value check. One of the most important techniques is identifying conjugate pairs in stereo images which are correspondent to the same point in the scene. The accuracy of matching algorithm affects the accuracy and stability of measurement. There are two requirements on matching algorithm, one is high precision and reliability, the other is high efficiency. This paper presents two methods to improve vision coordinate measurement according to the two requirements. First, valid measuring field significantly limits the searching space for finding conjugate pairs. It can also detect mismatches lying beyond the field and eliminate them. The second one is using geometrical and logical constrains regarding geometry model of spatial matching to detect and get rid of mismatches , to improve the accuracy of measuring results. The experiments demonstrate the good performance with the consuming time linear to numbers of points.4. Registration and fusion of 3D range data calculated in different view point When applied to wide viewing field or complex objects, vision coordinate measuring system can’t cover whole surface in single measuring process. To ensure the measuring space and precision, we divide measuring space into several subspaces and carry on measuring separately. Finally, integrate surface description are achieved by Tegistering separated surface data. This paper presents a method of constructing virtual points to compute transformation matrix using linear equations.And the last three parts are the emphasis of this paper.更多还原