Research On Preprocessing Technology Of UAV Lidar Data

Li DAR(Light Detection And Ranging)is a new detection technology developed rapidly in recent decades.It emits a single-band laser beam and obtains the 3d point cloud of the objects on the basis of the echo signal,and because of its high angular resolution and strong anti-interference ability,Li DAR has been widely used in the fields of remote sensing,surveying and mapping and environmental protection.Among all kinds of Li DAR equipment,UAV(Unmanned Aerial Vehicle)Li DAR system is popular among engineers and researchers due to its advantages of low cost,convenient in acquisition and ability to acquire large scale scene.However,there are many errors and noises in the original point cloud collected by the UAV Li DAR,leading to the data not up to the standard of research and practical use.Moreover,UAV Li DAR is a relatively new field,and few researches on its data preprocessing methods and workflows can be used for reference.All problems above seriously restrict the development of UAV Li DAR.To this end,this paper designs a data processing workflow specially for the UAV Li DAR acquisition system used by our research group.Starting from the original point cloud data acquired by the UAV Li DAR,the high-quality point cloud is the final output after the data processing workflow.This research is mainly composed of the following three parts:The first part deals with the problem that the point cloud precision decreases due to the integration error of the UAV Li DAR system.Firstly the equipment used in the experiment and the UAV Li DAR system of our group are introduced,and the data used in this research are described in detail along with the data acqusition plan.Next,starting from the system integration error model,this part analyzes the concrete manifestation of these errors on the point cloud.The system integration error calibration matrix is further solved by selecting the point cloud data with good GPS quality and using ICPatch(Iterative Closest Patch)algorithm to solve the transformation between the adjacent strip,and the matrix is then applied to the point cloud calculation process to acheive the Li DAR system integration error correction,resulting in the realative errors between multiple strips' point cloud are only several centimeters.The second part mainly solves the problem that the accuracy of the point cloud decreases due to all kinds of error aliasing in the acquisition process of UAV Li DAR system.Starting from the error model of UAV Li DAR system(excluding the integration error),this part analyzes the causes of all kinds of errors and their specific effects on the point cloud.By referencing the data-driven strip adjustment model,ICPatch and NDT(Normal Distribution Transform)algorithm are successively used to register the point cloud data after the system integration error calibration,so as to eliminate the error aliasing and improve the relative accuracy of the point cloud to several centimeters.Through the comparison with the commercial software Ri PROCESS,it is proved that the experimental results of the data processing workflow designed in this research are almost of the same with Ri PROCESS,so that the data processing workflow can be used as a substitute for Ri PROCESS in eliminating the errors in the point cloud.The third part mainly solves the problem that all kinds of noises in UAV Li DAR system lead to the degradation of the point cloud data quality.In this part,point cloud noises are firstly divided into outlier,duplicate points and coarse points according to the specific manifestation of the noises.Different kinds of filtering apporoaches are then carried out considering the causes of the noise—outliers are filtered by the method based on statistics,distance threshold based filtering approach is carried out for making duplicate points more uniformly distributed,and the coarse points then be smoothed by MLS.After the outlier removal,the rendereing of the point cloud elevation comes back to normal,and no outlier cluster remained.Then,by homogenizing the point cloud,the differences between the point density in several sample area are reduced--the standard deviation of point density is reduced to 25%-50%,and the overall point cloud turns to uniformly distributed,along with the high-density redundant point band extinguished.Finally,the point cloud was smoothed so that the root-mean-square error of the distances between the nearby points to their fitted plane was reduced to 20%,and the error was only centimeter-level,and the point cloud structural features were more prominent than before.To the end of the processing workflow,high-quality point cloud data are outputted,which can be used for subsequent application and research.