Research On 3D Building Model Reconstruction From Airborne LiDAR Point Cloud Aided By Aerial Image

Airborne Li DAR(Light Detection and Ranging)is a new and efficient technology for directly acquiring three-dimensional(3D)spatial data of ground objects,the point clouds obtained by airborne Li DAR has been widely used in smart city construction,object interpretation and other fields.As an important basic data of smart city and 3D geographic information system,the 3D building model reconstruction has always been a research topic and difficult problem in many fields.Airborne Li DAR point cloud is one of the important data for 3D building model reconstruction,and contains reliable 3D surface information of buildings,however,airborne Li DAR point cloud has the characteristics of irregular distribution,uneven density,and lack of semantic information,making it difficult to ensure the geometric accuracy of outlines of the 3D building models reconstructed by only utilizing airborne Li DAR point cloud for reconstruction.Aerial image has the advantages of significant spectral features,high spatial resolution,and rich semantic information.Taking advantage of aerial image to assist airborne Li DAR point cloud for 3D building model reconstruction is conducive to obtaining 3D building models with higher precision.Therefore,the paper focuses on 3D structured building model reconstruction from airborne Li DAR point cloud aided by aerial image.The main achievements and innovations are as follows:1.The research background and significance of the paper are expatiated,and the current research status and existing problems of airborne Li DAR point cloud classification,roof segmentation,building outline extraction and 3D building model reconstruction methods are summarized and analyzed,respectively.2.Aiming at the problems of low classification efficiency and requiring a large amount of training samples for airborne Li DAR point cloud classification with deep learning,an airborne Li DAR point cloud classification method based on deep residual network(DRN)transfer learning is proposed.The method introduces transfer learning into airborne Li DAR point cloud classification,designs a new point cloud feature image generation strategy,constructs and trains a new neural network classifier.The classification result is obtained after post-processing.The classification experiments on airborne Li DAR point clouds from multiple regions acquired by different types of sensors demonstrate that the method can improve airborne point cloud classification accuracy while reducing the training time and training samples.3.A roof segmentation method based on region growing(RG)and random sample consensus(RANSAC)is proposed.By utilizing the designed iterative region growing algorithm and RANSAC to extract reliable roof patches and small area roof patches,respectively,and then optimizing the extracted roof patches,the method effectively overcomes the influence of inaccurate segmentation caused by roofs which differ significantly in terms of their area,shape complexity,and number,as well as the uneven density and irregular distribution of the airborne Li DAR point cloud.Multiple representative building point clouds are utilized for segmentation experiments,it is verified that the method can obtain roof segmentation results with high accuracy,and has the ability to reliably segment the roofs of buildings with small area and complex structure.4.An airborne Li DAR point cloud building outline extraction method using neighborhood direction distribution is proposed.Building boundary points are extracted by the designed neighborhood direction distribution-based algorithm.By constructing a triangulation irregular network(TIN)and operating on the edges,boundary points can be tracked without set any length threshold,and an orderly extraction result of building outline can be obtained,which solves the problem of poor applicability due to the difficulty in parameter setting of existing building outline extraction method.Experiments were performed using simulated and real building point clouds with different densities and shapes,which demonstrates that the method can obtain building outline extraction results with high accuracy,and can also effectively overcome the problem of difficulty in parameter setting,and has strong applicability.5.A method for 3D structure building model reconstruction from airborne Li DAR point cloud aided by aerial image is designed.Multiple types of 3D building feature lines,e.g.,ridges and step lines,are generated based on aerial image and airborne Li DAR point cloud.Through the designed multi-feature line-based perception grouping algorithm,aerial image is integrated into the reconstruction process,thereby the advantages of airborne Li DAR point cloud and aerial image are effectively used to reconstruct 3D structured building models.Reconstruction experiments using building point clouds from two regions have verified that the method can effectively use the building outline information extracted from aerial images to assist 3D structured building model reconstruction from airborne Li DAR point cloud,and can reconstruct buildings with different level of roof structure complexity,and the reconstructed 3D structured building models are of high accuracy.