Extraction And Reconstruction Of Single Building Based On Dense Image Matching Point Cloud

As a city’s infrastructure,buildings are an important part of the city.The development of smart cities requires the data support of a single 3D model of the building.The traditional reconstruction of a single building adopts manual methods,which has the problems of low processing efficiency and high reconstruction cost.It cannot meet the needs of smart cities.The UAV tilt photography technology provides a new way for large-scale building reconstruction,but the building model reconstructed by the UAV cannot be operated alone.Using point cloud to extract and reconstruct the building unit can obtain a better unit model.Compared with the traditional Li DAR point cloud,the acquisition cost of the dense image matching(DIM)point cloud is lower,but the DIM point cloud has a lot of noise and low accuracy.Therefore,it is necessary to conduct high-efficiency and high-precision building monomer extraction and reconstruction research on low-cost DIM point clouds.This paper takes DIM data as the research object,focusing on the extraction of individual buildings,the extraction of building outlines and the research of building geometric reconstruction.The main research results are as follows:(1)Constructed a combined method of extracting individual buildings from DIM point cloud.Filter the ground points through the cloth simulation filter(CSF)algorithm;then combine the green index(EXG)and neighborhood information to remove the vegetation points;finally use the DBSCAN density clustering algorithm to segment the single building point cloud from the remaining point cloud.The experimental results show that this method can effectively extract the point cloud of individual buildings.Compared with the EXG method,the completeness of the buildings extracted by this method are higher.(2)Aiming at the high noise and low accuracy of DIM point cloud,a method for extracting the 2D contour of a single building is proposed.By intercepting the point cloud of the middle elevation of the building,project it to the xoy plane.Considering the low density of noise points after projection,perform density denoising to obtain building contour point clouds;then,combine random sampling agreement(RANSAC)algorithm and European clustering algorithm to classify contour points and fit contour line segments;finally,According to certain criteria,the contour lines are merged and sorted to obtain the 2D contour of the building.The experimental results show that this method can effectively extract the 2D contour of the building.Compared with the key points of the contour manually extracted by EPS,the plane point error is within 0.15 m.(3)Considering that the traditional plane segmentation method does not perform well in segmenting the DIM point cloud,a plane segmentation method based on contour constraints is used.On the basis of segmenting the building facade,the window point cloud is segmented using brightness threshold and rasterized;then the image processing technology is used to extract the windows,and finally the 2D contour lines are stretched to realize the geometric reconstruction of the building.The experimental results show that this method can get a better geometric reconstruction effect of buildings.