Research On The Total Least Squares-Based Registration Method For LiDAR Point Clouds

Due to the limitation of the field of view of LiDAR sensors,in order to obtain LiDAR point clouds that can comprehensively characterize the surface of the target geographic entities,it is usually necessary to set up several stations along different directions to collect point cloud data,however,the point cloud acquired by above station are relatively independent,thus point cloud registration is needed to achieve the fusion of point cloud from different stations and ensure the integrity of the geometric structure of the target entity surface.The Gauss-Markov model-based LiDAR point cloud registration algorithm is the most widely used,however,there is a problem that Gauss-Markov model describe the error of registration primitive is inadequate,which cannot guarantee the asymptotic unbiasedness of the solution.Based on the above analysis,this thesis replaces the Gauss-Markov model with the EIV model,establishes point cloud registration parameter solving models based on point,linear and planar features respectively,and solves the registration parameters under the criterion constraint of total least squares,mainly research work is as follows:1)Using the point feature as the primitive of LiDAR point cloud registration,Rodrigue matrix and the unit quaternion are used to describe the spatial rotation transformation,and the three-dimensional spatially similar transformation model based on the EIV model is established,objective function is constructed with the constraint of minimize the deviation of conjugate points from base station and the station to be registered,and the solution model of the registration parameters is constructed by iterative solution.In addition,considering that the iterative solution model depends on the initial value and has a low convergence rate,matrix decomposition is used to achieve the direct solution of the registration parameters.2)Taking the linear features in 3D space as the primitive of the registration,aim at the problem that the spatial linear fitting algorithm based on Gauss-Markov model fails to consider the error which along Z-axis,the spatial linear fitting algorithm based on the EIV model is given to realize the accurate extraction of spatial linear features of LiDAR point cloud.On this basis,Plücker coordinates are introduced to represent the spatial linear features,the spatial rotation transformation of linear features is described by Rodriguez matrix,and the solution model of LiDAR point cloud registration parameters based on linear feature constraints is established.3)Taking the planar features in 3D space as the primitive of the registration,for the problem that the EIV model fails to consider the structure of the coefficient matrix and the weight of the observations,a planar feature fitting algorithm based on the PEIV model and the Gauss-Helmert model is derived to realize the precisely extraction of the spatial planar feature.On this basis,based on the spatially similar transformation of the normal and modulus of the planar feature,quaternions are introduced to describe spatial rotation transformation,and the solution model of the registration parameters based on the planar feature is established.In order to verify the feasibility and correctness of the above LiDAR point cloud registration parameter solving model,the simulated data and measured data are used to test respectively,and the parametric solving results are compared with the conventional algorithm.The experiment shows that: Compared with Gauss-Markov model,EIV model can well take into account the errors of the conjugate features extracted from the base station and the station to be registered,the registration parameter solving model is more rigorous in theory and the registration parameters calculated are more accurate,which effectively enhances the binding force of LiDAR point cloud registration and improves the quality of point cloud registration.This thesis has 13 figures,20 tables and 69 references.