Research On Technologies Of Segmentation-Based Classification And Building Reconstruction From Airborne LiDAR Point Clouds

Airborne Light Detection and Ranging(Li DAR)is regarded as a new opportunity for fast acquiring and updating of spatial data,and it has been widely applied in the field of Geoinformatics.Nevertheless,despite the fact that most of the technical hardware difficulties and system integration problems have been solved,the development of algorithms and methods for processing of Li DAR data is still in the stage of exploration,and many problems need to be solved urgently.In this doctoral dissertation,point cloud segmentation,segmentation-based filtering,segmentation-based classification,and segmentation-based reconstruction of 3D building models are the key subjects for post-processing of Li DAR data.As a result,based on the point cloud segmentation,an object-based point cloud analysis concept is proposed,and it is employed for point cloud filtering,point cloud classification,and reconstruction of 3D building models from point cloud.Concluding,the main contents and innovations are listed as following:1.The positioning principle and composition of airborne Li DAR are briefly introduced,and the hot and key problems of point clouds filtering,point clouds classification and point clouds reconstruction of 3D building models are analyzed and summarized,providing the essential theories for engineering applications and follow-up algorithm design.2.The classic plane-growing-based point cloud segmentation method has the disadvantages about deficiency of large number of parameters,complicated calculation and frequent attempts.Thus,a new point cloud segmentation method for plane growing and smooth surface growing is proposed based on the constraints about normal and distance.In the process of region growing,the geometric distance from a neighboring point to the seed plane and the angle difference between a neighbouring point and the seed point are taken as two similarity measures for surface growing to achieve the segmentation of point clouds.Moreover,the competitive mechanism is put forward to impro ve the accuracy of the segmentation results.The above segmentation algorithm is further divided into planar surface growing and smooth surface growing.Experiments show that the planar surface growing method is suitable to cluster the man-made objects such as buildings and roads,and the smooth surface growing method is suitable to cluster the natural ground surface.Meanwhile,the above point cloud segmentation is the first step of the successive processing of the point clouds.3.Be aimed at the forested terrains with scarps and cliffs and the regions with complicated feature structures,a filtering method is proposed by merging of progressive TIN densification(PTD)filter and smooth surface growing segmentation.The smooth-surface-growing-based point clouds segmentation method is embedded into the PTD filter,which enforces the authenticity of the initial ground model.It is able to overcome the shortcomings of classic PTD.Experimental results suggest that,compare with the classic PTD,the improved method is capable of preserving the ground points nearby the terrain structure lines and the regions with complicated feature structure,which will significantly improve the accuracy and quality of point clouds filter.4.Based on the plane-growing-based point clouds segmentation,an object-based classification method is proposed for classifying the airborne Li DAR point clouds in urban areas.And the optimal feature selection of point clouds and the intelligent classification method are studied.The object-based classification takes the plane segment as the basic unit for processing and analysis.It is to calculate 10 features about the geometry characteristic,radiometric characteristic and echo characteristics.The RBF-SVM is utilized to classify the segme nts,and connected component analysis for 3D point clouds is proposed to optimize the original classification results.Experiments suggest that our proposed method is capable of making a classification of the urban point clouds with high classification accuracy,and the classification accuracy is promoted.5.An object-based method for reconstruction of 3D building models from Li DAR point clouds is proposed in urban areas,including t hree innovations:(1)Based on the above results of point clouds classification,TIN-based connected components analysis is put forward to separate the building point clouds into individual buildings,which realizes the individual buildings separation;(2)The outlines are automatically extracted for each building,and this step takes the 3D geometry element as a unit.(3)The inner structural lines are automatically extracted for each building,and this step takes the 3D geometry element as a unit.Meanwhile,a semi-automatic method is put forward to extract the elevation jump lines of complicated buildings and the attached objects on the building roofs.Experiments suggest that the proposed method is able to reconstruct the contour model of irregular and complex multistory building.Therefore,the proposed method is feasible for reconstructing the most buildings in urban areas.