The Study Of Multi-view Point Cloud Registration Method Based On Total Least Squares

The terrestrial laser scanning technology is more and more widely used in all walks of life,with its advantage of fast speed,high precision and non-contact measurement in recent years,and multi-view point cloud registration is a core technology in the point cloud data processing of terrestrial laser scanning.With the development of society and the advancement of modern surveying and mapping technology,the precision requirement of multi-view cloud data registration in all walks of life is also increasing.The registration accuracy is actually depend on the solution of the coordinate transformation parameters between the two stations.The paper proposes a new algorithms of multi view point cloud registration,and deduces the method and calculation process detailedly,based on the research of several major multi view point cloud registration algorithms.And the algorithms are implemented by MATLAB language,for the evaluation and comparison of the accuracy,The main research contents are as follows:The paper deduces the solution process of ICP algorithm based on singular value decomposition,by studying the iterative closest point algorithm,and elaborates the basic principle and method of the 3D Delaunay point set search method detailedly,which based on the 2D Delaunay triangulation method,and deduces the Delaunay-ICP algorithm for quickly searching and matching closest points,optimizes the time efficiency of ICP algorithm.The experimental results show that the Delaunay-ICP algorithm has the same parameter solution results and precision of ICP algorithm,but it's computational efficiency is better than the traditional ICP algorithm's,which increased by about 27.94%of ICP algorithm.Additionally,the defect of ICP algorithm,which is not suitable for the larger rotation angle between the stations,is verified.The Delaunay-ICP algorithm and ICP algorithm has a preferable effect in the cace of some smaller rotation angle between the stations,but can not find the correct corresponding points and cause a wrong registration results,when the rotation angle is larger.On account of the classical least squares and nonlinear least squares method,the paper deduces the basic solution of the linear model and nonlinear model of multi view point cloud registration,and verifies the solution by different case.The experimental results show that the linear model has the same high precision and better effect as the nonlinear model when the rotation angle is small.However,when the rotation angle is large,the linear model will be distorted seriously and must be solved through the nonlinear model.The paper also builds the Gauss-Helmet model based on Error-in-variables model and corresponding total least squares estimation,and deduces the solution steps in detail based on the nonlinear Lagrangian method of Newton-Gauss iteration.The measured experimental results,which acquired from terrestrial laser scanning,show that the proposed algorithm is suitable for any rotation angle of multi view point cloud registration,and has higher registration precision than ICP algorithm and nonlinear LS algorithm.Compared with the existing TLS method,the accuracy is consistent,but the number of estimated parameters could be greatly reduced,the computational is efficiency improved.In view of the mass point cloud registration problems,the paper proposes the Delaunay-TLS algorithm,a new multi-view point cloud registration method,which based on TLS and in the premise of the 3D Delaunay point set search strategy.The new algorithm uses the Delaunay point set search method to quickly locate the corresponding point set,and overcomes the problem that the Delaunay point set search method does not apply to the large rotation angle,by performing a coarse registration before performing large area point cloud precise registration.In addition,the new algorithm makes full use of the mass point cloud for precise registration,and has higher precision than traditional TLS algorithm.The experimental results show that this multi view point cloud registration algorithm is suitable for arbitrary rotation angle,and has higher registration accuracy compared with the TLS algorithm through a few feature points of stitching,especially for scanning the object feature points are extracted inappropriately.