Abstract: A novel registration method for point clouds is presented, which consists of three main steps. Corresponding points are found through the invariants of Gaussian images. Some erroneous coincidence relationships are eliminated based on the clustering characteristic of rigid transformations between corresponding points. The rigid transformation of the registration is obtained in an iterative process by using singular value decomposition. Experiments show that the invariants of Gaussian images are more effective in identifying corresponding points than curvatures. Compared with the traditional iterative closest point (ICP) algorithm, the proposed method is better to register partially overlapped point clouds.

Key words: C-space clustering, Gaussian image, Invariant, Registration, Singular value decomposition