Building Classification And Structural Line Extraction In Urban Scene From MLS Point Clouds

The buildings classification and structure lines extraction based on 3D point clouds are important research contents of 3D city modeling,and play an important role in the fields of smart city construction,urban information construction and urban planning.Mobile Laser Scanning system is the main means to quickly acquire high-precision 3D point cloud data in complex scenes.However,most of the point cloud data directly obtained has problems such as large data volume,uneven density,occlusion loss and noise,which brings great challenges to the processing of point cloud data.Challenges mainly reflect in the following aspects:Firstly,The large amount of point cloud data leads to the low efficiency of classification processing time;secondly,the uneven density and occlusion will result in incomplete classification and extraction of building objects;thirdly,the extracted building line structures is scattered,and there is no correlation and regularity,which can not be well applied to subsequent modeling research.Aiming at the above problems,this paper starts with the point clouds pretreatment,and improves the efficiency of classification and extraction while retaining the structure details of the segmentation point clouds.The accurate classification of building point clouds is realized by using the local context relationship between the point clusters.By analyzing the structural characteristics of building,and using the hierarchical relationship of geometric structures to realize the extraction and regularization of building structure lines.The research contents and results of this paper are as follows:(1)Over-segmentation of point clouds considering structural detail integrityBased on the improved over-segmentation method of VCCS_KNN,the original point clouds are segmented,which ensures the integrity of the structure details of over-segmented point clusters,and reduces the number of point cloud in classification processing.(2)Research on buildings extraction using the method of point clouds context classificationBased on the over-segmentation point clusters,the best distinguishing ability of point clouds features are extracted and selected.Based on MRF model and Graph-cut algorithm,the optimal neighborhood system between point clusters is constructed to obtain local context information,reduce the error propagation of over-optimization,and achieve the classification and extraction of global consistency of building objects.(3)Research on building structural lines extraction using geometric structure hierarchy relationBased on the analysis of the semantic and geometric characteristics of building structure,the geometric hierarchical relationship of building facade structure "plane-contour-line" is put forward,and the structural lines of buildings are extracted hierarchically.Through extracting the geometric consistency constraints between lines,the line structure is regularized,and the structural lines of buildings satisfying the requirements of modeling are obtained.Combining with the research content,this paper uses the measured MLS point cloud data in a small town and the open source Oakland point cloud data to verify the proposed method of building point cloud classification and structure lines extraction.The experimental results show that the proposed method can extract the complete building objects efficiently and accurately;the method based on the hierarchical relationship of geometric structure can extract more regular structural lines,which accords with the basic structural characteristics of buildings,and can be better applied to the research of 3D modeling.