Research On Three-dimensional Point Cloud Automatic Registration Algorithm

With the rise of digitization and informationization,3D scanning technology has also been continuously developed.The 3D point cloud data obtained based on 3D scanning equipment has been widely used in many fields such as robotics,medical image processing,cultural relics protection,and scene modeling.However,due to its limitations and external factors such as lighting and occlusion,the 3D scanning device cannot obtain the complete point cloud of each surface of the object to be measured at one time.It is necessary to scan the object to be measured multiple times from various perspectives to obtain "one-sided" point cloud data.These "one-sided" point cloud data are not in the same coordinate system.You need to use the point cloud overlap area to perform coordinate transformation to unify them into the same coordinate system.This process is called the point cloud registration.Point cloud registration is a key step in 3D reconstruction technology,and the quality of its registration results will affect the effect of the entire model reconstruction.Therefore,3D point cloud registration has once become the focus of research in 3D point cloud processing technology.Many research results have been achieved,but the existing algorithms still have problems in terms of speed,accuracy,robustness,stability and so on.In view of this,this paper further studies the point cloud without color information and the point cloud with color information based on the existing algorithms.The main research contents are as follows:(1)For the three-dimensional point cloud data without color information,in order to solve the problem that the classic ICP registration algorithm is prone to fall into local optimization,a secondary registration scheme combining coarse registration and fine registration is adopted.In coarse registration,feature histogram combined with RANSAC algorithm is used for coarse registration.In order to speed up the search for nearest neighbors,the KD-tree structure is first introduced into the algorithm.In order to further ensure the effectiveness of the rough registration algorithm,the key feature extraction part of the algorithm is improved so that the algorithm can effectively extract feature points without losing a lot of key information that is not obvious.In order to ensure that the algorithm has certain robustness,after obtaining the initial registration point pair,the rigid distance constraint and the RANSAC algorithm are used to further eliminate the mismatched point pair.However,in actual applications,only coarse registration does notmeet people's accuracy requirements for registration.It is necessary to further accurately register after coarse registration to meet actual applications.The fine registration algorithm proposed in this paper is optimized based on the existing multi-resolution registration point algorithm.First,the maximum resolution of the point cloud is obtained by the density of the feature points,and the sampling method of the feature points is optimized.This algorithm improves the accuracy of registration when the features are not obvious.It has been verified by standard point cloud data experiments that the method can effectively complete registration for point cloud data of different sizes and point cloud data with different degrees of noise,and its accuracy,speed and robustness have been improved.(2)For the three-dimensional point cloud data with color information,in order to make full use of its RGB color components,a new 4D-ICP algorithm is proposed.This method can solve the problem that the traditional 3D-ICP registration effect is poor when the point cloud surface is relatively flat and the geometric features are not obvious.In this method,the weighted average method is first used to convert the RGB values into gray values,and the weight factor is set according to the gray value variance and the sum of the curvature variance.Then adaptively adjust the influence of gray information and curvature information on the registration according to the weighting factors,and realize the organic combination based on gray information and curvature information.Through the verification of real experimental data,this method achieves stable registration of different point cloud data,and its accuracy is relatively higher than that of the existing 3D-ICP.