Abstract: Registration of point clouds is a key problem in model acquisition and reconstruction. A novel method is presented to register a pair of clouds by estimating the 3D moment invariant feature. In this method, the 2D moment invariant is expended to 3 dimensions, and it is used to describe the local surface characteristic feature of point clouds. With this feature searching for corresponding points, the method implements the iterative closest point algorithm (ICP). The algorithm calculates the characteristic description of point clouds, and every point in one cloud finds another point which has similar description in close field of other cloud as a corresponding point, then corresponding points set can be built by corresponding points. The set should be transformed to centroid coordinates for origin, and then covariance matrix can be established based on the corresponding points set and applied by singular value decomposition (SVD), the rotation matrix and translation matrix would be obtained from the result of SVD. Do the above steps until the iteration convergence. The registration result of two face clouds shows this method is feasible and effective.

Key words: 3D moment invariant, Corresponding points, Iterative closest point, Point cloud, Registration