Research On Point Cloud Registration Method Considering Plane Features

In recent years,terrestrial laser scanner have been widely used in many fields and have played an increasingly important role in the information age.The hardware of terrestrial laser scanner has been rapidly developed,but the development of data processing has a certain lag.Terrestrial laser scanner employs the station coordinate system,and the coordinate system of the data collected by the adjacent station is not uniform.Therefore,it is necessary to splice the cloud to unify the data coordinate systems of different stations.Point cloud registration is the first step in point cloud data processing,and the precision of which affects the accuracy of subsequent data processing.Therefore,point cloud registration is a crucial step of data processing.The dissertation focuses on the registration of point clouds,and the research work is mainly carried out from three aspects: rough point cloud registration,point cloud segmentation,and point cloud fine registration.The main innovations of the paper are as follows:(1)An automatic point cloud registration model is proposed with introducing RANSAC algorithm based on the automatic matching of plane features.This model utilize RANSAC algorithm to realize the automatic matching of datasets and model-focused planar features then the automatic registration of point clouds coming from two neighbor scan station are realized.Three planes are randomly selected from the data set and the model set.The plane parameters are used to calculate the coordinate system transformation parameter,which will be utilized to transform the data set.The angle between each plane in the data set and each plane in the model set is calculated and its threshold is set.The two planes whose plane angle is smaller than the threshold are the corresponding planes.Then,coordinate plane conversion parameters are calculated using the corresponding plane parameters.The experimental results show that the proposed algorithm can robustly and precisely realize the automatic registration of point cloud where the corresponding planes in adjacent stations are completely overlapped,partially overlapped,or not overlapped.(2)The model of point cloud segmentation algorithm based on RANSAC algorithm is improved.Compared with the traditional RANSAC algorithm,this algorithm increases two constraints.The first one is the parallelity among the normal vectors of the sample points.Three points,the angles among the normal vectors of which are less than a threshold,are randomly selected as the sample to calculate plane parameters.The second one is the parallelity among the normal vector of the interior point and the normal vector of the sample points.The interior point is selected when the angles among a point and sample points are smaller than a threshold.Experimental results show that the improved RANSAC algorithm reduces the judgments times of internal point set and improves the accuracy of plane regions extraction.(3)Based on ICP,a registration model is proposed taking into account the overlapping regions.Using the improved RANSAC algorithm to segment the point cloud of overlapping regions,and the point cloud is divided into plane point cloud and surface point cloud.Taking full advantage of the overlapping area of surface information,different methods are used to establish the corresponding points in different planes,which ensures the distribution of corresponding points in the entire overlapping area.Experimental results show that the proposed algorithm improves the point cloud registration accuracy of ICP algorithm and expands the scope of ICP algorithm.