| With the development of terrestrial 3D laser scanning technology,3D modeling based on ground 3D laser point cloud data is becoming more and more widely used.In order to acquire the 3D point cloud model,it is necessary to register the multiple viewpoints and perform 3D modeling on the registered point cloud data.Point clouds registration is the key technology of 3D modeling,in order to meet the needs of 3D laser point cloud registration and modeling,this paper analyzes the existing registration algorithm at first,then improves the algorithm of 3d normal distribution registration,analyzes the point cloud data compression method and establishes the model point cloud compression algorithm based on vector-angle method,moreover,conducts modeling research.Finally,an example application and analysis is carried out.The paper mainly completed the following four tasks:(1)According to different types of point cloud data,the corresponding registration algorithm is established.During the initial registration,the local feature initial registration algorithm based on SAC-IA algorithm and PFH/FPFH feature description operator,and the global search initial registration algorithm based on RANSAC algorithm and 4PCS are established respectively.The initial registration algorithm of point cloud suitable for different types of data is verified by experiments.On the basis of the initial registration results,the three-dimensional normal distribution transformation algorithm is improved to solve the problems of easy falling into local optimum and poor convergence.The improved algorithm uses the approximate Hessian matrix to solve the problem that the computation complexity of the three dimensional normal distribution transform algorithm leads to low registration efficiency,and the linear search algorithm is used to improve the Newton iterative algorithm to speed up its convergence.(2)Aiming at the problem of too much 3D point cloud data and too long modeling time,a point cloud compression algorithm based on vector angle method is established.Firstly,the point cloud data is partitioned by the bounding method,and the point cloud density in the small bounding box is used as the iteration threshold.At the same time,the point cloud data in the bounding box is iterated for many times,so as to partition the small bounding box of the adaptive point cloud data with different sizes.Then,the vector-angle method is used to compress the data of each bounding box,which ensures that the point cloud data is compressed greatly on the basis of not losing the features.(3)In this paper,the compressed cloud data are modeled and compared.Firstly,the point cloud compression algorithm established in this paper is used to compress the two groups of data,and compared with the common compression algorithm for point cloud compression efficiency.In addition,the Delaunay triangle-based modeling method is used to model the compressed point cloud.Eventually,the modeling accuracy is compared.(4)This paper uses the above theory to study the two different types of point cloud data.Experimental results show that the local feature initial registration algorithm based on SAC-IA algorithm and PFH/FPFH feature description operator,and the global search initial registration algorithm based on RANSAC algorithm and 4PCS have completed the point cloud initial registration.And the initial registration algorithm of point cloud based on FPFH feature description operator and the initial registration algorithm of 4PCS point cloud have good initial registration efficiency.During the initial registration on the basis of using the improved three dimensional Gaussian distribution transform algorithm of point cloud registration accurately,the experimental results show that the algorithm has obtained the better and faster registration,verify the effectiveness of the algorithm.The data compression modeling experiment of the point cloud data after registration also shows that the vector-Angle method has good applicability. |
PDF Full Text Request