Research On Construction Method Of 3D Point Cloud Real Scene Model Based On UAV-Mounted Mobile Lidar Scanning System

The 3D point cloud real scene model can completely move the earth into the computer,and different users directly acquire the required information on the computer to reuse data.It can technically solve the current multi-survey integration strategy proposed by the country.The 3D point cloud real scene model can provide data support for refined management needs in smart cities.The key to obtaining a complete 3D point cloud model is the fusion of multi-source data.The point cloud and image data are fused to generate a true color point cloud,which can give full play to the unique advantages of image and point cloud.The integration of point cloud data from different sources can build a complete point cloud real scene model with no missing and no dead ends.The thesis uses the Unmanned Aerial Vehicle(UAV)-mounted mobile lidar scanning system(ULS)as the main method to acquire data.In this thesis,according to the characteristics of the ULS sequence image and point cloud data,a fast registration fusion method of point cloud and image is established;In view of the lack of the generated 3D point cloud real scene model due to the characteristics of ULS technology,this thesis proposes a two-step repair strategy,which lays the technical foundation for the construction of a complete point cloud real scene model.The specific research contents are as follows:(1)First,using SURF with FLANN algorithm to achieve fast and accurate matching of the Intensity image generated from original point cloud based on Delaunay interpolation and the corresponding image collected by the airborne camera.Then use the collinear equation and the least square method to calculate the registration parameters.Finally,according to the registration parameters,the GNSS time index is used to realize the fusion of the sequence image and the point cloud,and a preliminary real scene model is obtained.(2)Aiming at the lack of point cloud and holes in the model,a two-step method of point cloud repair strategy is proposed:a)Use supplementary measurement method for the seriously missing areas to obtain a more complete point cloud model;b)Use the boundary shrinkage method to repair the smaller holes in the model to obtain a complete 3D point cloud real scene model.(3)According to the registration and fusion and point cloud repair methods proposed in this thesis,the experimental field is selected for verification.Experiments show that the fusion accuracy of the point cloud and the image is high.the root mean square error(RMSE)in the fusion reaches 0.026m for the 30-meter high data set and 0.029m for the 50-meter high data set.The accuracy of the patched overall point cloud model was evaluated.The RMSE in the model point is 0.059m,and the error in elevation is 0.063m.