Building Facade Structure Extraction Method Based On Imagery Laser Point Cloud Considering Semantic Information

With the progress of society,digital cities,digital tourism,smart cities,and smart transportation are developing rapidly.Buildings as places for human residence and activities,the demand for digitalization and modeling is constantly increased.The traditional manual modeling technology cannot meet the building model construction of the gradually expanding city.The three-dimensional laser point cloud,as an emerging spatial information data,has the advantages of high precision,high sampling rate and compact structure,which is widely used in building deformation monitoring,building fine three-dimensional modeling,building plots and ancient building restoration.However,the large and disordered volume of three-dimensional laser point cloud data can lead to problems such as computation time and data redundancy.The building facade structure,which includes edge segments and wall segmentation lines containing outdoor stairs,rain canopy,doors,windows,balconies,eaves,and pillars,provides key information needed for most building model reconstructions with a small amount of data.Therefore,extracting the building facade structure information from the laser point cloud of building facades is currently a major research hotspot.A method of image laser point cloud building facade structure extraction considering semantic information is proposed in this paper through the research and summary of existing methods.This method first performs region growth on the building point cloud,segments and clusters the building point cloud according to the plane,and obtains the building facade point cloud model containing different building facades.Then,considering the semantic information and image information,the building facade edge line segments are extracted as the building facade structure,and finally the extraction results are optimized to obtain an accurate and complete building facade structures.The specific research results are summarized as follows.(1)Segmenting and clustering building facades laser point cloud.Calculating the normal and curvature of each point through principal component analysis,and defining the similarity between points.The points with the same or similar properties around the seed point are merged into the region where the seed point is located.Then,the new points continue to grow around as seeds until no more point clouds can be included,thus clustering the building point clouds according to the plane and confirming the facades of the building point clouds.(2)Building facade structure extraction.The building facade point clouds are projected according to the plane,and generating a two-dimensional color image.Then,the building facade point cloud model,two-dimensional color image,semantic information and transformation matrix are combined to construct the building facade image point cloud data model.Finally,the Semantic Color Edge Drawing(SemColorED)algorithm,which is optimized based on the Color Edge Drawing(ColorED)algorithm,is used to extract the building facade structure from the building facade image point cloud.(3)Optimization and evaluation of building facade structure.The extraction results of building facade structure are optimized by using three steps: morphological stacking line optimization,main direction filtering and curve structure optimization.A comprehensive evaluation system was established,which evaluated and analyzed the experimental results of the proposed method and the comparative methods,verifying the effectiveness and superiority.The measured data and open-source data are combined to obtain the point cloud data of two buildings in university using the Riegl VZ-1000 three-dimensional laser scanning system,and a building data is cut from the Semantic3D open-source point cloud dataset as the test dataset.The results show that the building facade structure is extracted completely,and through quantitative evaluation,the superiority and robustness of the proposed method are proved.