Research On Accurate Registration Technology Of Three-Dimensional Point Clouds

As an important data-processing step in spatial three-dimensional point cloud preprocessing,point cloud registration plays an important role in obtaining an intact surface of the measured object with three-dimensional coordinates.However,the performance of the current point cloud registration algorithms remains to be improved in terms of speed,accuracy and reliability.In this paper the precise registration technology of spatial three-dimensional point cloud data including feature extraction,coordinate conversion,accurate matching and multi-station adjustment is mainly studied.The main work and innovations are as follows:1.The development of three-dimensional point cloud data registration is summarized,and it is divided into five parts,including pre-processing,feature extraction,initial registration,accurate registration,and multi-station point cloud adjustment.2.The spatial distribution rules of point cloud data are analyzed,and then an algorithm for extracting virtual feature point of point cloud data is proposed.The virtual feature points are extracted by the endpoint or straight line fitting algorithm,which is based on the clustered curvature feature points.The experiment results show that the virtual feature points have higher reliability,which can help to improve the calculation speed and accuracy of initial registration.3.To solve the problem that the traditional Bursa model and Rodriguez Matrix algorithm is susceptible to gross errors in initial registration,a Rodriguez Matrix algorithm based on Total Least Squares is proposed,which can take the error of the coefficient matrix into account,and effectively reduce the influence of the coordinate error of the feature point in parameter solution.Experimental results show that this algorithm has higher precision and robustness than traditional algorithms.4.Considering that the traditional registration algorithm has a limited search space and is easy to fall into local optimum,a two-step-based 3D-NDT registration algorithm is proposed.Compared with the ICP algorithm,this algorithm not only improves the accuracy and reliability significantly,but also has a higher calculation speed.5.A strip adjustment model based on quadratic or cubic polynomial is used to reduce the error accumulation of multi-station point cloud registration.The experiment results show that this method can effectively reduce the system error accumulation and improve registration accuracy.