Registration of Botanic Tree Point Cloud Based on Pseudo Feature Point

Registration for tree point cloud presents a high registration error due to the complex structure of trees and serious self-shielding. The paper proposes a registration algorithm based on pseudo feature point. This algorithm includes two steps. In initial registration, we use pseudo feature points to adjust the position of two original point clouds quickly and roughly at first. However, pseudo feature points sometimes can't fully represent the feature of original point cloud owing to the noise, it leads to a high registration error obtained in initial registration, and then need to use the improved sparse iterative closest point algorithm to adjust two original point clouds again. Experiments show that the proposed algorithm can register both non-leafy tree and leafy tree. Compared with iterative closest point registration and sparse iterative closest point registration, the method significantly reduces the registration error by 41.1% and 16.8% respectively under the same number of iterations. The method can also register nonplant point cloud.