3D Modeling Of Point Cloud Data From Laser Scanners

Mobile robot is an essential embranchment in robotics. With the development of mobile robots, a mobile robot becomes more and more intelligent and self-reliant. As an intelligent entity, a mobile robot should have these fundamental functions, such as perception, reasoning, decision making and implementation. The application of a laser scanner to a mobile robot is an extention of its perceptive capability.In this paper, registration algorithms of point cloud data, which are obtained with the laser scanner of a mobile robot, are investigated. As the point cloud data are recorded with a changing reference coordinate system as the mobile robot is moving, object and environment modeling is very challenging. Among the algorithms to reconstruct the model of an object or an environment with the scanning point cloud data, the iterative closest points algorithm (ICP) is widely used. But due to the number of the scanning date is very large, the registration efficiency is degraded, and the feature point dislocation is caused. Herein, an ICP algorithm, which uses three feature points in initial regestration, is presented to improve the classical ICP algorithm. Experimental results are presented to prove that the proposed algorithm can improve the speed of registration and reduce the error rate, moreover it avoids local minimum.The main content of this study is organized as follows,Firstly, an algorithm to process the scanning point cloud data is introduced in detail.Secondly, a method to obtain the feature point from the scanning point cloud data is proposed based on the derivation and the limit methods of discrete data, as these feature points are the points at the interface of two lines, or on the boundary contour of an object, etc. Here three crucial and representative points are used in the initial registration as reference points.Thirdly, the modified ICP algorithm is proposed based on the three feature points of the initial registration, and comparison and analysis is made with the classical ICP algorithm.The registration algorithm is implemented for 2D point cloud data first, and then is extended to 3D point cloud data. In the modified ICP algorithm, three calibrated feature points are used in the initial registration, and the classical ICP is then employed to realize accurate registration. In addition, kd-tree neighborhood search algorithm is used to improve the searching speed. The modified algorithm increases the registration accuracy and reduces the research time. A assessment model is introduced to analyze the the registration results. Crust algorithm is further proposed to realize the Delaunay Triangulation and modeling of point cloud data.