Research On Extraction And Application Of Line Structures From Laser-scanned Point Clouds

3D laser-scanning measurement technology has become one of the irreplaceable techniques for acquirement of 3D spacial information.Up to now,many different types of laser-scanning measurement systems installed on different platforms and suitable for different targeted scenes have been developed,including types of close-range,terrestrial and mobile.And corresponding point cloud data processing has also become a hotspot issue in related interdisciplinary field.Because of the "blind scanning" measurement style of mainstream equipments,points in point clouds are usually scattered distributed,so that no other information could be understood from it by computer except for three-dimisional coordinates.So,features of interests should be descripted through data processing,in which semantic description of line features is one of the most important aspects.Consequently,studies on extraction of 3D straight line features from 3D point clouds have also achieved great improvements.However,due to the fact that point clouds acquired by 3D laser-scanning measurement technology have the properties such as massive volume,scattered distribution of points,diverse types,as well as uneven precision and lacking of completeness,existing methods on 3D line feature extraction still remain the following disadvantages:(1)high computational complexity or low computational efficiency;(2)lack of adaptiveness for different data quality of point clouds;(3)low automatic processing ability;(4)lack of false control and accuracy assessment.So it is greatly important to exploit efficient,automatic,self-adaptive and high reliable 3D line feature extraction methods to promote and expand the applications of 3D laser-scanned point clouds.In this doctoral dissertation,3D straight line features are defined as mainly four types:interior structure line segment,exterior boundary line segment,texture line segment and geometric straight line.And research of this dissertation mainly focuses on the description of such types of line features.The main contents are as follows:1.Based on self-adaptive octree voxel generation and voxel-based region growing,an efficient method for planar feature segmentation from point clouds is proposed,in which most correlated thresholds are selected through statistics of voxel information and the region growing strategy for plane segmentation is voxel based instead of point based.In this method,point clouds are firstly voxelized in initial octree voxel width,and the geometrical features of each voxel are calculated,including normal vector,eigenvalue,three dimensionality features,etc.Then,the terminal constraints for octree subdivisions are determined through statistics,and a list of octree voxels with inhomogeneous sizes can be achieved after subdivisions of the initial octree under these constraints.And finally,planar features are extracted through voxel-based region growing in different levels under corresponding statistical threshold constraints.Experimental results show that the proposed planar feature segmentation method is adaptive to different kinds of laser-scanned point clouds and can get relatively fine planar features with rather high operating efficiency.2.An efficient method for extraction of interior structure line segments through voxel-based region growing is proposed.Aiming at dealing with problems remaining in existing algorithms such as low efficiency and reliability,this method applies voxel based nearest neighbors searching and region growing to speed up the process for recognition and segmentation of the distributing regions of interior structure line segments.Corresponding line segments are extracted in these segmented regions in the end and the accuracy of the results are assessed.Experimental results show that this proposed voxel-based region growing method for extraction of interior structure line segments has high accuracy and efficiency as well as the ability of achieving relatively more ideal results.3.A method for extraction of detailed edge line segments based on graphical projection of point clouds is proposed.To increase efficiency,3D point clouds dimension is reduced to 2D image dimension by graphical projection of pre-segmented 3D planar point clouds to avoid point based nearest neighbors searching.To preserve edge details,the projection resolution for each plane is determined through accurate statistics of average local intervals between points in voxels composing corresponding planar point clouds.To deal with some errors caused by uneven densities of planar point clouds,a structural element self-adaptive to local point density for mathematical morphological operation is applied for crack repairment of projected images.As for extraction of 3D edge line segments,back-projected 2D edge line segments extracted from images are utilized as references for clustering of edge points,and a 3D line fitting algorithm based on recursive weighted least squares is proposed to reduce the affect of edge noise.And finally,to control the quality of extracted 3D edge line segments,a series of refinement strategy is presented.Experimental results show that this proposed graphical projection based method for edge line segment extraction has high efficiency and fine processing results as well as automatic processing ability,which satisfies the efficient processing demand of large scale point clouds.This method provides important promotion to the extraction of integrated 3D edge line segments and its results have good application expectation.4.A method for extraction of texture line segments based on texture mapping images is proposed.On account of low reliability of existing related methods,a quality assurance and quality control strategy is utilized.In the first place,the quality assurance procedure is organized starting with the calibration of installation parameters between 3D laser scanner and external installed digital camera,which includes calibration of intrinsic parameters and extrinsic parameters of the digital camera.And finally,the quality control strategy through 3D texture line fitting of support points is adopted and a false control method based on three-dimensional normalized cuts is proposed.Experimental results show that this proposed method for extraction of texture line segments with a quality assurance and quality control strategy is effective,and is adoptable for scenes containing large numbers of foreground and background overlapping regions and uneven distributed densities.5.An automatic solution for rotation axis line extraction from 3D scanned point clouds of rotational-symmetric object is proposed.Given that existing methods have problems such as non-objectiveness,poor reliability and low adaptability,an automatic solution is proposed,in which almost all the surface points and their normal vectors are utilized as constraints.Firstly,the normal vector of each point in the point cloud is calculated and unreliable points are eliminated according to standard deviation values of local planar fitting for normal vector calculation.Then,the initial value of rotation axis is achieved through planar and spherical fitting of reference points which have same latitude with the randomly selected reliable seed point.Finally,refined result of rotation axis is calculated by solving the objective function listed according to the relationship between normal vectors of each reliable point and the rotation axis.The test experiments are performed,and the accuracy and precision of the method are verified by simulated and measured data.The experimental results indicate that the deflection degree is under 0.003° and the transverse distance is under 0.02mm,which satisfies the requirements of rotation axis extraction of rotational-symmetric object.This doctoral project relies on the support of Natural Science Foundation of China,and dedicates to automatic and efficient processing of 3D point clouds acquired by different types of laser-scanning equipments installed on different platforms for different targeted scenes.To be specific,semantic description of different types of straight line features is researched systemically,which provides important promotion to the extraction of line features from 3D point clouds.