Research On Multi-view Point Cloud Registration Algorithm In Complex Occlusion Area

When using a 3D laser scanner for mapping work,due to the influence of visual dead zone,the scan is divided into several areas for scanning measurement,which makes the measurement inefficient.In this paper,it is proposed to solve the problem by using the point cloud registration method of point cloud data collected from two different perspectives from the difference between ground 3D laser scanning point cloud data and aerial oblique photogrammetry point cloud data,and the theoretical results will improve the point cloud registration accuracy in complex areas,so that it can be widely used in 3D reconstruction,remote sensing mapping,unmanned driving and other fields.However,when performing point cloud registration on the measured data collected from the two perspectives,there are problems such as a large amount of point cloud data,a long time for the point cloud registration algorithm,and low point cloud registration accuracy.In view of the above problems,the research content and research results of this paper are as follows:(1)Aiming at the problem of low efficiency of point cloud registration algorithm caused by the huge amount of point cloud data,this paper proposes a multi-view point cloud registration method using CDF feature points.This method makes full use of the two advantages of 3D laser scanning to collect data from the ground perspective and oblique photogrammetry to measure from the aerial perspective,realize the integration of aerial and ground perspectives,and effectively reduce the impact of occlusion on surveying and mapping work.In this paper,the point cloud data of a building is used for registration and fusion,and the accuracy of the point cloud data obtained after fusion is verified,and the experimental results show that the registration error and registration efficiency are improved when the registration method is used.Moreover,the point cloud measurement accuracy completed by this method meets the requirements of mapping accuracy,which is an effective mapping method for occluded areas.(2)To address the problem that the traditional sampling consistent initial alignment algorithm has low alignment accuracy and is not efficient.In this paper,we design a method for joint measurement in heterogeneous multi-view point clouds using an improved sampling consistency initial alignment algorithm.The point clouds acquired by the 3D laser scanner from the ground view and the densely matched point clouds acquired by the tilt photogrammetry from the air view are aligned and fused using the improved Sample Consensus Initial Alignment(SAC-IA)algorithm.In this paper,experiments are conducted with the point cloud data obtained from 3D laser scanning of a building and the dense matching point cloud data obtained from tilt photography,and the accuracy of the point cloud data obtained after alignment fusion is verified.The experimental results show that the alignment accuracy and efficiency of the algorithm are significantly improved compared with the existing methods when using this paper.(3)Aiming at the problem that the classical SIFT algorithm still has a high proportion of redundant points when extracting point cloud feature points,this paper proposes a CDFSIFT(Curvature Density Feature SIFT,CDFSIFT)feature point algorithm,and uses the algorithm to carry out point cloud registration experiments.Firstly,the curvature density feature operator is introduced to improve the SIFT algorithm,secondly,the algorithm is used to extract the point cloud feature points to fully express the source point cloud data,and then the feature point is used to calculate the fast point feature histogram to comprehensively describe the feature point information,and finally the sampling consistency initial registration method is used to realize the coarse registration of the point cloud and the iterative closest point(ICP)algorithm is used to complete the point cloud fine registration.Experimental results show that the algorithm has good stability and can well express the shape and contour of the source point cloud.Based on the algorithm in this paper,point cloud matching is carried out on time,and the registration efficiency is increased by more than 40%.