Filtering And Segmentation Of Lidar Point Cloud Data

In recent years,with the continuous improvement of the ability to obtain 3D information,people are more and more in-depth research on 3D data.Compared with the traditional two-dimensional image,three-dimensional data contains more information with spatial characteristics and object characteristics.Three dimensional data can be obtained in many ways,among which lidar is widely used in various fields of environmental perception,such as unmanned driving,because of its advantages of high resolution,strong robustness and not affected by ambient light.Therefore,it is of great significance to study the analysis and processing methods of laser point cloud data.In the current research,the key of LIDAR point cloud processing technology is to select and reconstruct the target model from a large number of point cloud data.How to improve the efficiency of point cloud data processing and ensure the authenticity of the model through the research of denoising and segmentation algorithm is a hot issue.To solve this problem,this paper studies the processing methods of laser point cloud data,including:(1)This paper introduces the composition of lidar system and the principle of laser imaging,analyzes the data characteristics and noise of laser point cloud,and puts forward a denoising method of combining statistical filtering and bilateral filtering for laser point cloud,in which statistical filtering is used to remove a large number of outliers in the laser point cloud,aiming at the noise introduced by the system itself and the environment of the target object during the scanning of lidar Point and bilateral filtering smooth denoise the laser point cloud on the premise of maintaining the characteristics of the point cloud,and carry out voxel grid filtering for a large number of point clouds,while maintaining the outline of the point cloud,simplify the number of point clouds,and adopt the kdtree structure as the topological relationship of the laser point cloud,so as to speed up the filtering speed of the point cloud and improve the data processing efficiency of the laser point cloud.(2)Because the segmentation effect of the laser point cloud is very important for the subsequent target recognition and map construction,after the laser point cloud filtering,for the point cloud data with a large number of ground point clouds,the random consistent sampling algorithm is used to extract the ground point cloud of the plane geometry model.For the outdoor scene with sparse distribution of laser point cloud,an improved adaptive Euclidean clustering algorithm is proposed,which makes the distance threshold change with the point cloud coordinates to achieve a better segmentation effect.For the indoor scene with close distribution or even overlapping parts,the hypervoxel is used to mark the concave convex relationship,and along the convex region The segmentation of growth can realize the accurate segmentation of laser point cloud.(3)The principle of lidar ranging is analyzed and the upper computer platform of point cloud is built.The data collected from lidar ranging and the 2D coordinates of MEMS galvanometer are combined into the 3D point cloud data of laser,and the visualization platform of point cloud is built by using the PC end.The lidar system has the functions of point cloud real-time display,coordinate transformation and so on.The point cloud filtering and cutting algorithm modules are added to realize the visualization of laser point cloud data processing,and the subsequent target classification and obstacles are realized Object detection and three-dimensional map construction provide a good platform and have a good prospect in practical application.