Research On Point Cloud Registration Method Based On Linear And Planar Geometric Features

The three-dimensional point cloud model of city buildings,as a kind of high-density true-color point cloud data,naturally presents the state of the city,which can solve the difficulties in city management systems,such as slow generation of three-dimensional data,poor currentity and difficulty in updating.Current city point cloud data collection methods include:airborne laser scanner system(ALSS),mobile mapping system(MMS)and Terrestrial Laser Scanning System(TLSS)and so on.Due to each system is limited by its own carrier platform,scanning method,and viewing angle.it is difficult to collect the object to be measured completely(especially buildngs).In order to obtain the complete 3D point cloud model of city,combine with their respective advantages,study the multi-source data registration fusion achieve complete and accurate description of the urban building model This paper analyzes the different resolution of point cloud data in each system and the difficulty in extracting points feature constrained.It conducts research on point cloud registration based on line and surface geometric features.The main research work are summarized as follows:1.Studied common theories of point cloud registration,such as nonlinear seven-parameter method and nonlinear thirteen-parameter method.These models require Taylor series expansion,initial values introduced for iterative calculation and the number of iterations and conversion accuracy affected by initial value selection.In order to solve the initial value problem,the Rodriguez matrix transformation model and the Quaternion transformation model are studied.The Rodriguez matrix can solve directly,but existing singularities.The quaternion-based conversion model can solve the above deficiencies and is suitable for large-angle rotation,simple parameter description and fast calculation.2.The method of point cloud registration based on the feature of the same name is studied.The rotation matrix R and the translation matrix T are solved by the corresponding geometric features such as points,lines and planes,to achieve point cloud registration.Due to the influence of the resolution of the data collected by the device,there is no correspond-ing point in the actual point cloud data,which affects the accuracy of point cloud registration.This paper based on the linear features and planar features constrained to achieve point cloud registration And verified by experiment,the both can get higher accuracy of registration.3.Combined with the experimental area,after a series of point cloud preprecessing,get the point cloud data that meets the requirement of registration.The experimental results show that the ground point cloud data registration accuracy within 5 cm,and airborne and ground point cloud registration precision also reached the higher registration.