Research On Point Cloud Data Processing Algorithms Based On Lidar

In recent years,with the rapid development of laser scanning technology,it has become a reality to obtain three-dimensional information of the target in a non-contact,high-density and digital way.Therefore,this technology is also called the real-scene reproduction technology.The emergence of lidar is a major breakthrough in laser scanning technology,which has flexible structure and efficient processing.And the collected data is usually presented in the form of point cloud,which is easier to store and process.When the lidar scans a large-scale real scenes,it usually needs to obtain complete point cloud data in multiple directions and angles.There is a lot of noise in the collected massive point cloud,which results in the subsequent three-dimensional modeling effect is not ideal.Therefore,point cloud data processing technology is very important in 3D modeling,mainly including point cloud filtering,simplification and registration.This paper takes the 3D point cloud data acquired by lidar as the main research object.The work of dissertation on the key technologies of point cloud data processing is as follows:1.In view of the fact that the bilateral filtering method cannot deal with the noise points far away from the main body of the point cloud,and the flat surface is easy to cause excessive fairing of the model.The paper proposes a filtering algorithm based on feature region division.Firstly,the source and mathematical model of noise in laser radar point cloud are analyzed,then the three-dimensional point cloud space is divided according to the average curvature.Subsequently,different filtering methods are adopted for flat region and characteristic region to make point cloud filtering more targeted.Finally,the simulation results show that the filtering effect of this method and the execution efficiency of the algorithm are better.2.In order to solve the problem that the curvature-based simplification algorithm may lead to over-simplification of the flat surface or even lose the feature information of the object itself.This paper adopts the point cloud simplification algorithm based on geometric features,introduces the normal vector into the curvature simplification algorithm.Then,the appropriate simplification rate is selected according to whether the local feature information is rich.The experimental results show that the algorithm avoids the “hole” phenomenon caused by oversimplification,and retains the original details of the target object.Also,it has a good simplification effect on point cloud model and large-scale scene point cloud.3.When the overlap area between the two point clouds to be registered is relatively small,the iterative closest point registration algorithm has the disadvantages of long running time,many wrong points,and slow registration speed.In view of the above problems,this paper first uses the principal component analysis method to make the initial registration of the point cloud data from different angles;and then uses the ICP algorithm based on the feature points to register the point cloud after the initial registration.The method reduces the probability of mismatching and finally achieves the purpose of accurate registration.Experiments show that the algorithm overcomes the shortcomings of ICP algorithm,such as poor registration effect and long timeconsuming.And the algorithm is suitable for the real scene point cloud registration.