基于多线激光条纹测量的匹配与重建研究

【作者】 刘文学；

【导师】 方漪；

【作者基本信息】 青岛大学 ， 计算机应用技术， 2007， 硕士

【摘要】 本课题提出了一种基于多线激光条纹的视觉坐标测量技术，并在该技术中运用外极线约束的方法对多线激光条纹进行有效匹配，重建三维物体。该技术改进了在被测物体上添加标记的测量手段，是一种非接触式快速获取三维物体外形的测量方法。该方法先采集包含物体表面形状三维信息的畸变激光条纹图像，之后对图像进行改进的差影法处理，再对这些图像进行预处理、阈值分割、条纹细化等一系列处理后，根据双目立体视觉相应的数学模型和图像处理算法对畸变条纹进行分析，其中对应条纹间的匹配采用基于极线约束的配策略，得到被测物体表面的三维外形数据信息。在此基础上，通过实验验证极线约束匹配的精确性和有效性，并利用OpenGL，对实验结果数据进行三维重构和点云处理。具体来说研究结果包括以下几个部分：1、多线激光条纹标记的产生。生成多线激光条纹标记时需要通过确定光源的适当的焦距、强度和组合方式，使所投射出的激光线具有较普遍的适用性。本文通过反复的实验，制作了一种适当的光源发生器件，并对其一些重要参数进行了比较准确的描述。2、标记图像的采集与处理。为了将激光条纹能从原始图像中提取出来，本文设计了适当的图像采集方案。采用改进差影法进行条纹提取，并且较好地解决了由于采集条件的不稳定性(环境光线的不确定性变化、图像采集器本身的不稳定性等)所产生的噪声干扰的抑制问题。根据标记图像结构简单的特性，采用Ostu方法对差影后图像进行有效分割，之后采用数学形态学细化方法对条纹二值图像进行细化，得到单像素宽的条纹图像。3、基于极线约束的条纹匹配及重建。本文的匹配方法提出了利用极线约束的理论，进行激光条纹间的匹配，并根据实验的实际情况提出了多线激光条纹的匹配算法，提高了匹配的精度。在匹配的基础上，对物体进行了重建。4、实验及数据可视化后处理。通过实验验证极线约束匹配的精确性和有效性，并在VC++平台上利用OpenGL对实验结果进行三维重构和点云显示，进行了数据的分组排序、重新采样、数据网格化，更好地逼近被测物体表面。更多还原

【Abstract】 The paper brought forward a vision coordinate measuring technology based on the multi-stripe laser marks. In this technology, correspondence between marks was effectively completed by epipolar constraint and then 3D object was reconstructed. This technology improved measuring method attaching marks on object, overcame shortcomings of being unstable and low efficiency that manual measuring means had, and it was a non-contact measuring method capturing figure of 3D object in celerity.Firstly, in the method, images with aberrant laser stripes containing information about figure of 3D object were captured, then improved subtraction of the images was used to deal with these images, then pretreatment, threshold division and stripe thinning were performed on them.Secondly, further analysis was performed based on mathematics model of binocular 3D vision and images processing arithmetic, in which the matching between stripes was fulfilled on strategy of epipolar constraint correspondence. Then the data of object 3D figure was obtained.Thirdly, experiments were carried out to validate the accuracy and validity of the method of epipolar constraint. By OpenGL, results of experiments were displayed and part of which were executed by dot cloud processing.Research results can be described concretely as following:1, The production of striped laser marks. The laser marks generator should be designed with proper focal length, intensity and combining means. After myriads of experiments, a proper laser mark generator was designed, and several important parameters were exactly described in the end.2, Capturing and processing of marked images. In order to extract multistripe laser marks from images, this paper presented proper image capturing project. The operation of improved subtraction of images was used to extract the stripe laser marks, which effectively solved the elimination of noise interference produced by unstable capturing conditions from light and cameras. For the simplicity of structure of laser stripes images, the Ostu method was effectively applied in image division. Then single pixel stripes were generated via thinning means of morphologic processing of image.3, Correspondence based on epipolar constraint and rebuilding. Matching method in this paper was brought forward on the theory of epipolar constraint to match different stripes. According to practical instances of experiment, improved epipolar con- straint matching arithmetic was used in the paper and matching precision was heightened. Based on this, object was rebuilt.4, Experiment and results visualizing. Experiments were carried out to validate the accuracy and validity of the method of epipolar constraint. On the platform of Visual c++, by OpenGL, the results of experiments were displayed and part of which were executed by dot cloud processing, grouped sorting, re-sampling and data grid, so that the object surface could be approached better.更多还原