Building Extraction Based On Airborne LiDAR Point Cloud

Building extraction based on airborne LiDAR point cloud is of great significance for urban scene analysis,urban 3D modeling and urban map updating.Common buildings in urban scenes include houses,walls,street facilities and more.LiDAR point cloud can provide accurate spatial coordinates and three-dimensional spatial information.When using airborne LiDAR in urban scene analysis and urban map updating,it is necessary to use three-dimensional semantic segmentation to segment multiple building categories under the scene.When using airborne LiDAR in urban 3D modeling,it is necessary to distinguish between independent buildings such as houses in the scene,and three-dimensional instance segmentation is needed to extract each independent building.However,the airborne LiDAR point cloud has the problem of complex background,and the obtained point cloud contains a large number of non-building points such as ground and vegetation.The complex background brings difficulties to the extraction of buildings.In view of the above background and problems,the main research contents of this paper are as follows :1.In order to improve the performance of point cloud segmentation in complex background,a feature enhancement module that can be used to improve the accuracy of 3D semantic segmentation is proposed.The module uses point coordinate embedding and self-attention mechanism to do feature enhancement.The module can fuse the features of low-level point cloud and pay attention to important features.It has obvious improvement in the difficult segmentation of buildings such as walls,bridges and street facilities.Comparative experiments were carried out on two public data sets,which both improved the segmentation accuracy,and ablation experiments were designed to verify the effectiveness of the module.2.A two-stage 3D instance segmentation method is studied to extract the most important building instance-houses in LiDAR point clouds.Aiming at the defects of similar methods in extracting houses at present,a three-dimensional instance segmentation method for house extraction is proposed,this method can remove the interference of background points on the prediction of housing clustering auxiliary information,and can more effectively extract the house and distinguish each instance in the house.The airborne LiDAR data set containing a large number of buildings was produced,and the comparison experiments were carried out with different 3D instance segmentation methods.The effectiveness of the method was verified by additional design experiments.