三维激光扫描点云数据配准算法研究

发布时间：2018-05-24 01:10

本文选题：点云数据 + 粗配准　；参考：《昆明理工大学》2017年硕士论文

【摘要】：三维激光测量方法凭借数据采集速度快、精度高的优点,在三维物体和三维场景的建模中得到了广泛的应用。然而由于坐标测量装置的视域范围限制和被测物体及场景本身的尺度以及周围环境的限制,一次扫描往往不能获取物体及场景的全部点云数据,因此需要进行不同视角下点云数据的配准拼接,以形成一个完整的场景。三维激光点云数据配准时,首先要推导出不同视角下点云数据之间的旋转平移变换关系,并将获取得到的不同视角下的点云数据统一到同一个坐标系下,这个过程就是实现对点云数据进行粗配准和精确配准。本研究针对传统的点云数据配准过程中存在着精度和计算效率不高的问题展开实验研究,试图改进现有的点云配准方法。论文对点云数据配准所涉及到的基本数学原理进行了阐述。将两个不同视角下的待配准点云分别作为源点集与目标点集,采用了基于快速点特征直方图(FPFH)描述子以及采样一致性方法实现点云数据的粗配准。先对点云数据提取关键点,计算关键点的表面法线,进一步利用法线特征计算快速点特征直方图(FPFH)描述子,然后利用采样一致性算法完成两片点云数据的粗配准。实验结果表明利用这一方法能够有效的优化点云数据的初始匹配位置。精确配准时利用粗配准得到的初始值,结合最近点迭代(ICP)算法来实现点云数据的精确配准。为减少点云数据的数据量以提高计算效率,引入了体素化网格法对点云数据进行了精简处理,再利用RANSAC算法进行错误匹配点对的去除来提高配准的精度,采用上述方法构造出精确匹配点对,利用先前研究计算出的优化初始迭代值,进行迭代,直到满足某个约束条件,最后完成两片点云之间的精确配准。与传统的最近点迭代算法相比,利用上述方法改进后的最近点迭代(ICP)算法在匹配准确度和计算速度上都有很大的提高。最后,结合PCL点云库,利用斯坦福大学点云数据库提供的bunny数据和dragonStand数据进行了实验比对,结果表明与直接利用最近点迭代(ICP)算法相比,本文提供的方法在增大匹配度和减少计算时间方面都优于传统方法。
[Abstract]:3D laser measurement method has been widely used in 3D object and 3D scene modeling because of its advantages of fast data acquisition and high precision. However, due to the limitation of the scope of view of the coordinate measuring device, the scale of the object under measurement and the scale of the scene and the surrounding environment, the whole point cloud data of the object and scene can not be obtained by a single scan. In order to form a complete scene, the registration of point cloud data from different perspectives is needed. When 3D laser point cloud data match punctuality, we must first derive the rotation and translation transformation relationship between point cloud data from different angles of view, and unify the obtained point cloud data under the same coordinate system. This process is to achieve rough registration and accurate registration of point cloud data. In order to improve the existing point cloud registration methods, this study aims at the problems of low accuracy and low computational efficiency in the traditional point cloud data registration process. The basic mathematical principle of point cloud data registration is expounded in this paper. Based on the fast point feature histogram (FPFH) descriptor and the sampling consistency method, the rough registration of point cloud data is realized by using two different point of view cloud as the source point set and the target point set, respectively, and the fast point feature histogram (FPFH) descriptor and the sampling consistency method are used to realize the rough registration of the point cloud data. Firstly, the key points are extracted from the point cloud data, the surface normals of the key points are calculated, and the fast point feature histogram (FPFH) descriptor is further calculated by normal features, and then the rough registration of the two pieces of point cloud data is completed by using the sampling consistency algorithm. Experimental results show that this method can effectively optimize the initial matching position of point cloud data. Using the initial value obtained by rough registration and the nearest point iteration (ICP) algorithm, the accurate registration of point cloud data is realized. In order to reduce the amount of point cloud data and improve the computational efficiency, a voxel mesh method is introduced to simplify the point cloud data, and then the RANSAC algorithm is used to remove the mismatched point pairs to improve the registration accuracy. The exact matching point pairs are constructed by using the above method, and the initial iteration values calculated by the previous studies are used to iterate until a certain constraint condition is satisfied, and the exact registration between the two point clouds is finally completed. Compared with the traditional nearest point iterative algorithm, the improved nearest point iteration (ICP) algorithm improves the matching accuracy and computing speed greatly. Finally, the PCL point cloud database is used to compare the bunny data and dragonStand data provided by the point cloud database of Stanford University. The results show that the proposed algorithm is compared with the nearest point iterative algorithm. The method presented in this paper is superior to the traditional method in increasing the matching degree and reducing the computational time.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P225.2

【参考文献】