Progressive Registration Of Ground Laser Scanning Point Clouds Integrated With ICP And NDT

With the rapid development of three-dimensional space information technology,the global laser radar market is becoming more and more larger.This poses greater challenges and requirements for automatic registration of point clouds.At present,scholars at home and abroad have carried out fruitful research on the methods and improvements of point cloud registration,but there are still shortcomings in the efficiency and accuracy of point cloud registration.And in the evaluation of accuracy and efficiency of point cloud registration,the evaluation methods are mostly self determined by the author,and there is still a lack of a more visualized and intuitionistic error distribution.The way of expression and the system of evaluation.Therefore,this paper improves the traditional ICP point cloud registration algorithm,proposes a GaussNewton algorithm to improve the nonlinear least square method to solve the point cloud transformation matrix,and integrates the 3D-NDT to achieve fine registration;constructs a systematic and intuitive evaluation system for the accuracy and efficiency of the point cloud registration;finally,it adopts indoor and outdoor points.Cloud scenes are compared with various point cloud registration algorithms.The main research work and achievements in this paper are as follows:(1)Improving the point cloud registration algorithm of traditional ICP.Before the registration is started,the K-D tree data organization structure is used to search the corresponding points of the two point clouds quickly and reduce the time consuming to retrieve the corresponding points.In the ICP algorithm,the solution of the target function is solved by using the nonlinear least square method of Gauss-Newton in the iterative target function,and the optimal solution is efficiently obtained.The transformation matrix.The improved registration and optimization of the improved ICP and NDT methods,through the PCA pre registration,through the selection of the appropriate parameters in the NDT process,complete the rough registration of the NDT method,and integrate this result with the refined Gauss-Newton registration method improved by Gauss-Newton to meet the requirements for accuracy and efficiency of point cloud registration.(2)A system evaluation method for point cloud registration accuracy and registration efficiency is proposed.In view of the problem that the point cloud registration evaluation method is not specific and systematic,this paper analyzes the related factors that affect the accuracy of point cloud registration and registration rate.In view of the inadequacy in the accuracy evaluation of point cloud registration,and considering the error distribution within the point cloud,the accuracy registration method based on the error ellipsoid is proposed,and the convergence is constructed.A comprehensive evaluation system for speed,iteration efficiency and iteration speed.(3)Registration test of indoor and outdoor point cloud scene.Taking the corridor scene of the 5 building of Chinese surveying and mapping innovation base and the indigo road section of Beijing as the case area,the registration test of indoor and outdoor point cloud scenes is carried out by the point cloud registration optimization algorithm proposed in this paper,and the evaluation method of accuracy and efficiency of the point cloud registration in this paper is made with the use of ICP alone and the use of NDT alone.In order to verify the accuracy and efficiency of the proposed algorithm and the wide applicability of the algorithm,the robustness of the proposed registration algorithm is verified by taking into account the influence of the overlap range and the movement of the space position.