Research On The Key Techniques For Building Roof Top Reconstruction From Aerial LiDAR Point Clouds

Since 3D building models are important framework data for the digital city,their acquisition issues have drawn numbers of attentions during the past two decades.The Airborne Light Detection and Ranging(LiDAR)system can directly generate the 3D position of the target objects and avoid the difficulties in stereo matching confronted by traditional photogrammetry methods,which makes it widely researched for building models reconstruction.Owing to the recent advances of data acquisition technology,increasingly dense and reliable point clouds can be produced.This provides the possibility to capture complex structures and subtle details,while also bringing significant challenges to current reconstruction methods.As a matter of fact,even when only considering the planar roof surfaces,the reconstruction procedures can still be challenging because of data insufficiency and building type complexity.The major difficulties exist in the following aspects:1)the scene complexity.It's still difficult to extra building points from a complex scene,thus the reconstruction procedures are still greatly affected because of the influence of none-roof points and the lack of data.2)the building complexity.The roof planes have various shapes and large range of scales,what makes it even tough is that the relationships among adjacent planes are too complex to be simply described by uniform rules or building primitives.As a result,the issue of over-regulation frequently occurs.3)the difficulties in features extraction and topologies construction.Due to the discrete nature of the point cloud,the mutual interference among roof planes,as well as the issue of occlusion,errors in roof segmentation or distinguished connections are inevitably produced.All those issues greatly limit the precision and completeness of the reconstruction results.In this work,we concentrate on the major issues confronted during the procedure of roof reconstruction,and our major contributions are concluded as follow:1)In allusion to the difficulties in facets features extraction,great efforts are drawn upon the issue of reliable roof segmentation.A weighted RANSA(RANdom SAmple Consensus)based segmentation algorithm is proposed,based on various improve approaches and strategies proposed in current works.Specially,to solve the issue of the spurious planes,which are widely suffered by RANSAC based methods,we thoroughly research the approach to improve the definition of the weight functions,so as to best reflect the inconsistence between inlier points and roof planes.The detail issues includes:first,the standard for the "best" weight function;second,how to evaluate different weight forms;third,how to involve the normal bias into the weight function to better suppress the spurious planes.Finally,we compare different weight forms and dedicate the "best" weight functions for the segmentation issues.2)In allusion to the difficulties in roof features distinguishing,we further research the robust extraction methods,including the generation of roof ridges,step edges,model corners and boundary edges and the regularization issue.Specially,we discuss the determination of regularization orientations under complex scenes as well as the integral adjustments of multiple features,which eventually make the balance between adaptability and regularization of the features.3)In allusion to the difficulties in roof topology distinguishing and representation by current methods,we propose a hierarchical roof topology structure for robust topology reconstruction.For traditional RTG(Roof Topology Graph)based methods,the thresholds are hard to be set when distinguish plane-plane relationships and leak or error connections frequency occurs.Our HRTT(hierarchical roof topology tree)solve those issues based on the following ideas:first,introduce the concept of building primitives or sub-structures in the topology structures and improve or constraints the roof topology by plane-model or model-model relations;second,we use a progressive topology construct strategy that detect robust connections preferentially.Those robust connections will separate the whole model into simpler components step-by-step and produce the basic semantic information for the identification of ambiguous ones.In this way,the effects from structures of minor importance or spurious ridges can be limited to the building locale,while the common features can be detected integrally;third,the extra verifying process and constraints are introduced to improve the roof topology and correct the possible errors.4)In allusion to the possible errors in constructed models,we design a robust model repairing tool.The method starts directly from the constructed models and compare them with the raw point cloud,then can then precisely locate the error regions of the models and correct them under a hypothesis-testing based approach.The proposed method can ensure the consistency between the constructed model and the raw point cloud and avoid holes in constructed models.