A novel robust point cloud registration method based on directional feature weighted constraint

Abstract

Laser scanning can obtain the three-dimensional point cloud data of the object surface in real time, which is widely used in 3D reconstruction, robot path planning and other fields. 3D point cloud data is essential for the reconstruction of geometric models, Point cloud registration is subjected to complex feature point search, mismatch rate and strict initial registration parameters. In this paper, a novel point cloud registration method based on directional feature weighted constraint is proposed. Firstly, voxel filtering and outlier removal are performed on the original input point cloud to reduce the computational complexity and improve the registration efficiency. Secondly, Random Sample Consensus(RANSAC) is used to lower the mismatch rate among the feature correspondences and to obtain more robust matching points. Thirdly, the neighborhood feature description of each feature point is constructed through the directional vector feature constraint among its neighbor points. Finally, the dynamic weights between key-points with robust adjacent features are calculated iteratively. The experiments are carried out on public and self-collected dataset. The experimental results show that, compared with the traditional ICP and NDT method, our method can significantly decrease the complexity of searching, distinctly accelerate the registration convergence speed, and improve the robustness and effectiveness of the point cloud registration.