基于轮廓线的自动三维重建方法研究

发布时间：2018-02-28 03:39

本文关键词： 轮廓线三维重建组件化控制点网视觉感知　出处：《青岛科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：基于单幅图像的三维重建是虚拟现实、计算机图形学等领域的重要研究内容,从单幅二维图像恢复出三维信息本身属于逆向工程,存在多义性的病态问题,可以用轮廓、亮度、纹理、明暗、交互、运动、变焦等方法进行三维重建。但是,重建效率低和精度不高仍然是研究中存在的核心问题。本文对单幅图像中正侧视图的物体进行建模,提出了基于轮廓线的自动三维重建方法。首先对输入图像进行预处理,然后根据物体的不同形态特征,选择使用对应的组件化建模算法,完成三维重建。主要工作包括:(1)详细阐述了基于轮廓线三维重建的预处理操作。通过比较Graph Cut算法和GrabCut算法的优劣,选择更优秀的GrabCut算法进行图像分割;提出了基于道格拉斯普克(Douglas Peucker)算法的距离插值方式的轮廓点重采样方法;采用快速细化算法提取目标轮廓的骨架,为三维建模提供有效的轮廓和骨架信息。(2)根据人类视觉感知原理,提出了基于控制点网的组件化建模算法,通过观察主体组件的形态特点,结合人体视觉感知,提出了控制点网建模的思路。首先综合考虑轮廓曲率,最近点、交点分别与轮廓点连线的长度和角度关系,选择构建最优点对,并在最优点对连线上设置控制点,经过连线构成二维控制点网;然后,以人类视觉感知为基础构建三维提升曲线,对二维控制点进行空间提升,进而对其进行优化和空间三角剖分,构建三角网格模型;侧肢组件由于结构和形态存在较大差异,又细分为对称流线型和一般结构两种类型分别建模。最后对主体和侧肢组件模型进行合并和纹理映射等操作来完成建模。实验表明,基于轮廓线的分组件建模方法效率高、建模效果好。(3)搭建3DCreator系统平台验证上述算法和技术。给出了多种物体的建模结果,并进行了功能和性能测试。系统集成了CGAL平台,可对重建后的三维模型进行细节优化、编辑等后处理。本文提出的基于轮廓线的自动三维重建方法,可为用户提供方便、快速、所见即所得的建模体验,支持三维打印、虚实交互、计算机动画等具体应用。
[Abstract]:3D reconstruction based on single image is an important research content in virtual reality, computer graphics and other fields. It is reverse engineering to recover 3D information from single two-dimensional image, and there is a morbid problem of ambiguity, which can be used for contour, brightness, etc. Texture, shading, interaction, motion, zoom and so on. However, the low efficiency and low precision of reconstruction are still the core problems in the research. In this paper, the object of the orthotropic image in a single image is modeled. In this paper, an automatic 3D reconstruction method based on contour is proposed. Firstly, the input image is preprocessed, and then according to the different morphological features of the object, the corresponding component-based modeling algorithm is selected. The main work includes: 1) the preprocessing operation based on contour 3D reconstruction is described in detail. By comparing the advantages and disadvantages of Graph Cut algorithm and GrabCut algorithm, the better GrabCut algorithm is selected for image segmentation. Based on Douglas Peucker (Douglas Peucker) algorithm, a method of resampling contour points based on distance interpolation is proposed, and a fast thinning algorithm is used to extract the skeleton of target contour. According to the principle of human visual perception, a component-based modeling algorithm based on control point network is proposed. The idea of modeling control point network is put forward. Firstly, considering the relationship between contour curvature, nearest point, intersection point and the length and angle of contour point line, the optimal pair is selected and the control point is set on the optimal pair line. Then, the three-dimensional lifting curve is constructed based on human visual perception, and then the two-dimensional control points are optimized and triangulated to construct a triangular mesh model. Because of the great differences in structure and shape, the lateral limb assembly is subdivided into two types: symmetric streamline and general structure. Finally, the main body and side limb component models are modeled by merging and texture mapping. The experimental results show that, The component modeling method based on contour line is efficient and the modeling effect is good. The 3D reator system platform is built to verify the above algorithms and techniques. The modeling results of various objects are given, and the function and performance are tested. The system integrates the CGAL platform. The automatic 3D reconstruction method based on contour can provide users with convenient, fast and WYSIWYG modeling experience, and can support 3D printing. Virtual real interaction, computer animation and other specific applications.
【学位授予单位】：青岛科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】