Fast 3D Modeling Of Buildings Based On Laserpoint Clouds

With the maturity of the technology of terrestrial 3D laser scanner, its high efficiency, rapid, high precision, large area to obtain object three-dimensional space distribution information of features are widely used in the field of digital city, the protection of cultural relics and building three-dimensional reconstruction. The focus of this paper is to extract the feature points from the point cloud data and realize the building’s fast, accurate and automatic reconstruction.The research object of this paper is the point cloud and image data acquired by the ground 3D laser scanner.First of all, it’sthree-dimensional reconstruction preprocessing, including the point cloud data mosaic and segmentation.In order to improve operational efficiency, we can segment the buildings into unilateral facade and do the following semantic recognition at the mean time.Then we adjust the attitude of semantic feature patch point cloud data to extract the edge points of the corresponding attribute space by block processing.The geometry frame of building wall is obtained by constructing the normal vector of wall plane and adjusting the normal parameters.By the method of combining the hole algorithm and the axial feature, the window boundary points are extracted and obtained by the normalization algorithm.According to the area of missing point cloud data seriously, we combine the image data as complement. We use RANSAC algorithm to fit edge points. The feature points coordinate are obtained by the normalization algorithm. We fuse feature point data and write feature points data into DXF file. Importing it into Auto CAD or 3DSMAX to map automatically and complete the automatic reconstruction of geometric model ofbuilding.At the same time, the conversion relationship between point cloud and image is used, and texture information is automatically captured by the feature points, and realized by manual processing.Finally, the quality of geometric model reconstruction is evaluated by texture information quality.Experimental results show that the method of program modeling is fast, automatic and accurate, and the feasibility of this method is proved by real data.