Research On 3D Laser Point Cloud Building Segmentation And Surface Hole Repair

In recent years,3D laser scanning technology has permeated various research areas.When acquiring 3D point cloud data,due to the limitations of the equipment itself,the different complexity of the target and the influence of environmental factors,the acquired point cloud data is disordered,uneven density and lack of semantics,and even point cloud holes,which seriously affect the accuracy of the model reconstruction.Therefore,it is important to make accurate segmentation and hole repair of point cloud data.The main research of this paper is as follows:(1)To address the problem of under-segmentation and over-segmentation of roof surfaces due to the complex spatial structure of building roofs,a segmentation method combining SIFT and 3D Hough transform is proposed.Firstly,the curvature calculation is introduced to improve the extraction degree of the SIFT algorithm,and then the Kd-Tree is constructed by using the feature points,searching five neighbouring points to fit the triangular surface,calculating its normal vector and putting it into the 3D Hough space,judging the reliability of the plane by the angle value between the normal vectors of the fitted surface,reducing the chance of pseudo-plane,and finally using the accumulator and peak detection to obtain the segmentation results.(2)To address the problem of over-segmentation and under-segmentation due to the low robustness of the seed points of the region growing algorithm.A segmentation method is combining FPFH feature grading and region growing is proposed.The Kd-Tree is first constructed,then the FPFH feature values are calculated and ranked,and the points with relatively low feature values are selected as the initial seed points to ensure the stability of the seed points,and finally the set of points is used as the base for region growing to segment the building facade point cloud.(3)An iterative slicing reconstruction method for point cloud surface holes is proposed to address the failure of traditional hole repair methods in repairing surface holes with uneven density and similar curvature variation.Firstly,the hole boundary is extracted,then the minimum enclosing box is constructed to achieve uniform segmentation,then the density of the segmentation result is used to judge the iterative slicing,and finally the moving least squares method is used to fit the slices and repair the missing part of the surface.(4)A point cloud hole repair method with value-added conditional moving least squares is proposed to address the problem of holes in the point cloud model due to the limitations of scanning equipment.Firstly,the hole boundary points are extracted,the point cloud model is iteratively sliced based on density analysis to preserve the local features,then the discrete group points are projected and quadratically fitted to ensure that the number of nodes is sufficient for hole repair,and finally the repair is carried out using the additional value-added conditional moving least squares method,and the repaired point cloud generated by the curvature constraint is used to realise the 3D reconstruction of the model.