Research On Feature Extraction And Application Of Buildings In Mobile Backpack Point Cloud

3D laser scanning technology with its advantages of high scanning accuracy,high speed and portability,can easily obtain millions of point cloud data on the surface of scanned objects,which has become an efficient technical means of spatial data acquisition.Therefore,speeding up the point cloud data processing speed to improve the utilization of data is one of the current research hot spots.However,most of the feature points are directly extracted from point cloud data by slice projection,triangular mesh construction,and single differential characteristics,which is affected by the point cloud density and local scene interference.On the basis of this,this paper proposes the method of point cloud segmentation and double threshold feature extraction based on RANSAC and CSF,to explore an automatic extraction method of feature points from 3D point cloud data,which suitable for common urban buildings and structures,the scanning object can be described with few and effective points.The results are as follows:(1)The 3D laser scanning system of mobile backpack and its advantages in data acquisition and processing are are compared and analyzed,and the original point cloud data are obtained after the trajectory and SLAM calculation of the scanned POS data by the backpack system in a teaching building and underground pipe gallery.(2)A total of 1526615 data points of one point cloud in the underground pipe gallery are selected as the test data of the algorithm,the RANSAC algorithm can effectively avoid excessive segmentation and insufficient segmentation.For the segmentation and extraction of the original point cloud data of the pipe gallery,by setting the threshold t,the sampling resolution and the minimum number of points,4649004 surface point cloud data containing only the overall contour information of the underground pipe gallery are segmented.(3)Because the teaching building point cloud data contains the information of the ground and ground features,the extracted building point cloud is a non-ground point,the conventional gradient filtering algorithm eliminates most of the ground point cloud data,but there are still some residual point cloud floating on the ground.By setting the mesh grid size,the maximum iteration number,the threshold that determine whether it is a ground point,split 6010353 point cloud data containing only above-ground buildings by CSF filter..(4)For the segmented point cloud data,the judgment points and their adjacent points are projected onto the micro section based on PCA algorithm,and the boundary points are extracted by comparing the angle threshold with maximum value of the adjacent vector angle difference formed by the identified surface point cloud.According to the best clustering index,the k-means clustering algorithm is used to cluster the normal vector to extract the edge points.It extracted 158274 feature points of pipe gallery,with the elimination rate of 96.09%;and extracted 369001 feature points of teaching buildings,with the elimination rate of 93.86%.(5)Statistics construct the relative accuracy of the 3D model and the original point cloud based on the extracted feature point.The mean square error in the 3D model data of the teaching building is 0.097 meters,and the mean square error in the3 D model data of the underground pipe gallery is 0.063 meters.The experimental results show that this method can effectively and quickly extract the feature contour targets in the point cloud data of building and structure,the extracted contour is clear and complete,which is in line with the original point cloud accurately.While preserving the contour features,the data is streamlined,which lays a foundation for the efficient utilization of 3D point cloud data.3D reconstruction based on the extracted contour point cloud,the data takes up less storage space because of removing a large amount of redundant data from the point cloud data,thus realizing the rapid and accurate reconstruction of the scanning object,the efficiency of the whole reconstruction process and the accuracy of the model are improved.