| Buildings are important features in urban construction,and play an important role in modern smart city construction,disaster prevention and control,military investigation and many other fields.Airborne LiDAR can quickly and accurately obtain coordinate information of 3D space.3d reconstruction of urban buildings based on airborne LiDAR point cloud data is of great significance.However,there are a lot of features in point cloud data,so there are some problems in efficiency and accuracy of building extraction and 3D reconstruction.Therefore,how to extract buildings accurately,quickly and intelligently is particularly important.This paper will use airborne LiDAR point cloud as the main data source and remote sensing image as auxiliary data to carry out building extraction and 3D reconstruction.The research work is as follows:(1)Domestic and foreign research status and theoretical basis introduction.This paper analyzes and summarizes the research status of building extraction from airborne LiDAR point cloud and remote sensing image.This paper introduces the characteristics of LiDAR point cloud data,expounds the workflow of BUILDING 3D reconstruction based on LiDAR point cloud,and focuses on the analysis of several key technologies of neural network and A-Shape contour extraction algorithm,providing theoretical basis for subsequent work extraction.(2)Building contour extraction based on LiDAR point cloud.The advantages and disadvantages of the progressive encrypted triangulation filtering algorithm and the progressive morphological filtering algorithm are analyzed and compared.The experimental comparison shows that the progressive morphological filtering algorithm is suitable for separating ground points from non-ground points of airborne LiDAR point cloud,and the overall error is only 4.4%.For non-ground points,the separation of low,medium and high vegetation is realized in Terrasolid software,and for separating high vegetation point cloud,buildings in point cloud data are extracted by building rules.For building point cloud,building contour is extracted by alpha-Shapes algorithm with single threshold.After experimental analysis,alpha-Shapes algorithm can extract building contour effectively.(3)Building edge detection optimization based on image assistance.Firstly,the advantages of building edge detection in remote sensing image based on deep learning are analyzed.In view of the limitation that LiDAR point cloud cannot accurately extract the contour under the condition of occlusion or small building space,the building contour extracted from remote sensing image by deep learning method is used to help optimize the building contour extracted from LiDAR point cloud.Through experimental comparison and analysis,u-NET network has better extraction effect,with Io U reaching75.2%,PA 86.04% and Recall 85.66%,all of which are better than Deeplab V3+ network model extraction results.For the plaque fragments existing in the semantic segmentation results,the threshold method was used to remove the plaque noises that did not conform to the characteristics of the building in the results.After the re-accuracy evaluation,it was found that the accuracy of IOU,PA and Recall indexes improved by nearly 10%.Finally,the building contour extracted from remote sensing image was further optimized to extract the building contour from point cloud,which improved the accuracy of building contour.(4)3D reconstruction of buildings based on key points.In this paper,a 3d reconstruction method based on building key points is proposed.The optimized building contour is regularized in the main direction,and the key points of the building contour are extracted as the bottom information of 3d reconstruction,and the roof key points of the building are extracted from the original point cloud by the regularized contour.Through the comparative analysis of experiments,it is found that the extraction accuracy of building key points can reach 85%.Compared with other 3D reconstruction methods,it is found that the keypoint-based 3D reconstruction method not only occupies less storage space,but also has high 3d reconstruction efficiency. |
