Research On The Method Of Urban Point Cloud Classification Based On The Vertical Structures Of Ground Features

There are several categories of city features,for example,water,plants,buildings and roads,all of which are important parts of the data of city information construction."3S"integrated technology are used to monitor the change and distribution of city features,and this is very meaningful to the city planning,management and construction.Therefore,in order to further the construction of smart city,researching a more accurate and rapid city features classification method is in need.Traditional photogrammetry,because of long production cycle,is far from meeting the demand of fast city information.Digital city rapid laser radar(LiDAR,Light Detection and Ranging)use active sensing system of 3D scanning technology,using laser as a carrier and the controlled light to reach the target on the ground,and it can accept the backscattering process from the ground target to accurately determine the distance.Because the LiDAR system need anything but illumination,and can obtain digital terrestrial 3D information without considering the time of the day.At present,in the digital city,LiDAR has become a common measuring tool,can directly obtain the measured three-dimensional coordinates of the target surface point cloud,with low cost,high speed,high precision,because of which it is widely used to obtain the city.Therefore,using LiDAR data as the basis for the identification and classification of surface features,can reduce data acquisition time,get accurate three-dimensional features of the object,and at the same time improve classification accuracy.Because the LIDAR point cloud is of high density,large size,discrete information,uneven distribution,so it cannot get surface texture information like image data,plus complex shape of the city spatial distribution,it is to extract feature directly from the point cloud.The algorithm efficiency and accuracy faces a great challenge.Based on a large number of research results,this paper takes all kinds of objects in different characteristics in the vertical structure as the basis of classification,at the basis of the vertical structure calculation of airborne LIDAR point cloud data,realizes the classification of city point cloud.The main research methods describes as the following several points:(1)Analyze the vertical structure characteristic of ground features,based on the distribution characteristics of airborne laser scanning point cloud,divide point cloud into grids in a certain scale,with a fixed height scale,slice horizontal grid into several levels.Based on the relationship of the ground vertical structure characteristics and features of point cloud vertical structure,calculate the value of normalized vertical structure for each horizontal grid.(2)Because in the vertical direction,the point cloud of each layer can calculate the corresponding vertical structure value,and form a feature curve,but not all features are effective for point cloud classification.Therefore,before the classification,the PCA method are used to reduce the dimension of the feature curve to remove similar characteristics of all curves,so that we can retain the characteristics of greatest difference,and improve the classification accuracy and efficiency.(3)Based on the dimension reduction of the feature curve,using K-Means and other classification algorithm for curve classification,the feature curve similar to each other are the same class,so as the point cloud in the grid,and finally the classification of the city point cloud is finished.(4)The results of analysis and accuracy assessment of classification.There are lots of variable in algorithm flow,such as the level of horizontal grid scale,vertical stratification scale,the effective number of dimension reduction,choice of vertical structures.Therefore,analyzing of the impact of these variables on the results,seeking for the optimal way of expression of the vertical structure characteristic curves,are unavoidable to obtain the best classification results.In the aspect of the classification accuracy,based on aerial image data and on-the-spot investigation,this paper compares the classification result with the artificial classification and use aerial image data in the field as a reference.In this paper,the method of urban point cloud classification is proposed based on the vertical structure of the ground objects,though plenty of experiments and accuracy evaluation,this paper verifies the feasibility of the proposed method.