The Automatic Identification Method Of Typical Features For The Large Buildings Based On 3D Laser Point Cloud

With the rapid progress of society,a wide variety of large buildings have been set up like the bamboo shoots after a spring rain,which has become an integral part of the modern city.However,in the process of building use and construction,it is very easy to occur the phenomenon of settlement and deformation in the building due to the influence of internal or external factors.This will bring a huge threat to the society,people's life and property safety.Therefore,it has very important practical significance to monitor the deformation of large buildings.As a new measurement technique,3D laser scanning technology has the characteristics of fast,high precision and non-contact.This make the 3D laser scanning technology widely used in different fields.But,the point cloud data obtained by using this technology has the characteristics of large amount of data,high redundancy and so on.Thus,how to extract the typical features of the point cloud data from the complex point cloud has become a hot research topic in deformation monitoring.As the most important geometric information of a building,surface features and line features accurately reflect the displacement and deformation of the building.In order to effectively extract the plane features and line features from the point cloud data of the building,this paper takes the point cloud data of the new library of South Lake Campus of Wuhan University of Technology as an example to carry out the experiment.Based on the theory of tensor voting and the Gaussian graph clustering,we focus on how to accurately extract the deformation characteristics of the building(plane and line features)to carry out the research,and carry out the experiment on Matlab software platform.The main research as the follow:(1)The extraction of face feature from point cloud data of large building.As one of the most important geometric features of a building,the surface features can be used to reflect the change of the displacement of each part of the building.In order to accurately extract the surface features of the building from the complex point cloud data,we propose a multi-scale tensor voting method based on tensor voting,and use this method to select seed points.Then,extracting the surface features of building based on the theory of region growing.(2)The extraction of line feature from point cloud data of large building.As the other most important geometric features of a building,the line features can reflect the changes in the overall situation of building displacement very well.However,the line feature of buildings has the characteristic of diverse.It will ignore the extraction of sharp feature boundary lines if we only use traditional topological relations.In view of this problem,we firstly to mark the line features of the point cloud data based on the topological relationship between the point cloud.Then,marking the line features with sharp features using adaptive Gauss graph clustering method,which is based on the correlation theory of Gauss graph clustering.The displacement of different parts of a building can be well reflected if the surface features and line features are applied in the deformation monitoring of buildings.Firstly,it can eliminate the influence of non-building points by extracting building plane information,and the number of point cloud is reduced at the same time,it will improve the computational efficiency and accuracy.Then,combined with the two kinds of line feature extraction methods to classify the different types of line features,it make up the shortcomings of traditional methods in the extraction of sharp features of the boundary line,and improve the accuracy of line feature extraction.