【作者】 罗敏；

【导师】 周春艳；

【作者基本信息】 中南大学 ， 计算机科学与技术， 2011， 硕士

【摘要】 随着地面、车载、机载激光扫描采集系统技术的成熟,国内外越来越多的研究人员开始研究基于激光点云的物体三维建模。点云特征提取作为基于激光点云三维建模的一个重要环节,也逐渐成为其中的一个研究热点。针对已有点云特征提取算法不易于理解、实现难度大、数学运算复杂等缺点,综合数字图像和激光点云各自的优点,本文提出数字图像辅助激光点云特征提取方法,并给出详细的方法流程及实验结果。核心思想是将点云中的点映射到对应的二维图像中的像素,然后从图像中提取目标物体的特征,将特征所包含的像素根据对应关系找到在点云中的对应点,之后对这些点进行曲线和曲面拟合,由此得到点云的特征。需要解决的关键问题就是点云和对应二维图像的对应关系的建立,即点云和图像的配准。为了区别,本文将点云对应的由CCD相机获取的图像称为CCD图像。利用强度图像能够反映原始点云大部分特征的特点,将点云与CCD图像同名点的提取转换为点云强度图像与CCD图像的同名点提取。针对CCD图像和点云强度图像灰度差异大而不能使用传统的基于灰度信息的配准问题,引入医学图像配准中常用的基于互信息的图像配准算法,针对特征点的提取使用改进的自适应阈值Harris角点提取算法。利用共线条件方程解决点云和CCD图像的配准问题,通过曲线拟合和曲面拟合实现点云特征的精确提取。实验结果表明本文方法能够比较准确地提取点云特征,并且实现难度小、没有大量复杂的数学运算,具有一定的应用价值。更多还原

【Abstract】 As the terrestrial, vehicles borne, air borne laser scanning data acquisition system technologies matured, more and more domestic and foreign researchers began to study on 3D modeling based on point cloud. Point cloud feature extraction as an important link in 3D laser scanning modeling, has gradually become one of the research hotspots.It is difficult to understand and realize the existed algorithms of point cloud feature extraction. To deal with these disadvantages, it puts forward a new algorithm to extract features from point cloud indirectly in this paper which named digital image assists point cloud feature extraction, and a detailed process of this algorithm and results of the experiment have been given. The core idea of this algorithm is mapping the points in the point cloud to the pixels in the 2D image,then get features of target object from 2D image, According to the registration relationship between the pixels in 2D image and points in point cloud, the points in point cloud compose the features can be found, using curve and surface fitting to make features more accurate, these curves or surfaces are the features extracted from point cloud. The key problem that needs to be resolved is the establishment of registration relationship between 2D image and point cloud. To avoid confusion, the 2D image acquired by CCD camera is called CCD image here. Since the intensity image of the point cloud can reflects most features of the original point cloud, the registration between point cloud and CCD image can be divided into two steps. First, register the CCD image and intensity image, in view of grayscale variation between CCD image and intensity image is too great to register them by use of traditional image register algorithms which based on gray information, the image register algorithm based on mutual information usually applied in medical image registration is introduced. Improved adaptive threshold Harris corner detection algorithm is used to extract feature points from images. Second, register the CCD image and point cloud, the collinear condition equation is introduced to solve this problem. Since registration relationship between CCD image and point cloud has been established, features extracted from CCD image can be mapped into the point cloud, with curve and surface fitting, the curves or surfaces fitted in point cloud are the features of point cloud. The experiment results show that this algorithm can extract features from point cloud more accurately and with less operations, can be applied in some solutions.更多还原