| Point cloud registration technology plays a vital role in projects such as computer-aided virtual restoration of Terracotta Warriors and Terracotta Warriors and Wisdom Museum.The quality of the registration can even affect the quality of the entire project.The digitization of cultural relics is the primary work of these projects.Due to the particularity of the terracotta warriors and the limitations of the currently used 3D scanner function,it takes multiple scans to obtain all the data of the terracotta warriors.To get a complete model of terracotta warriors and related cultural relics,the point cloud model in the local coordinate system needs to be unified under a coordinate axis and registered into a completed point cloud model.This process depends on the registration algorithm of the three-dimensional point cloud.The terracotta point cloud registration algorithm proposed in this paper is a registration algorithm based on feature space,including initial registration and fine registration.The main research contents include:(1)An initial registration algorithm based on the combination of the feature space and the difference between the main axis angle distance is proposed.First,the feature space of the feature point to be matched is used to select the matching pair.If the similarity of the feature space reaches the specified value,You can initially confirm the matching point pair;then use the principle that the difference between the main axis angle distance does not change before and after the position conversion,and select the initial matching point pair again;finally,rotate the registration point pair and translate the transformation to obtain the transformation parameters,and finally Complete the initial registration of the point cloud to be registered.The experimental results show that the initial registration algorithm based on the combination of feature space and the difference between the main axis angle distance has improved the efficiency and accuracy of registration in the point cloud for the initial registration task.(2)An improved ICP algorithm based on the combination of feature space and neighborhood feature search is proposed,which improves the accuracy of point cloud registration,significantly reduces the number of iterations,reduces the time taken for registration,and improves Efficiency of registration.First,select the matching point pairs through the feature space,and then use the principle that the corresponding neighbors of the matching points also correspond to each other to filter the matching point pairs again,and finally use the ICP algorithm to iterate to obtain the final fine matching.quasi.The experimental results show that the improved ICP algorithm based on the combination of feature space and feature neighborhood in the task of point cloud fine registration,in addition to the smaller matrix transformation error,the registration results obtained are more robust and stable.(3)Design and implement a point cloud registration system based on feature space.The system implementation realizes the import of terracotta warriors and horses data,to point cloud preprocessing,initial registration,accurate registration,and display of registration results.It is a registration system with relatively complete functions.The highlight of the point cloud registration system designed in this paper is that it takes less time and obtains more accurate results.The key technology to achieve this effect lies in the initial registration algorithm and fine registration algorithm proposed in this paper,as well as the best registration result achieved by the combination of these two algorithms. |
