A novel two-stage algorithm for accurate registration of 3-D point clouds

Accurate registration of 3-D point clouds is a common problem in computer vision. This paper presents a new two-stage algorithm for point clouds registration. A novel local invariant feature which is a k-dimensional vector is proposed and used in our coarse registration stage. Two new structural constraints combined with the Iterative Closest Point (ICP) algorithm are adopted in our fine registration stage. The accuracy and effectiveness of our algorithm is visually and quantitatively demonstrated by the comparative experiments on synthetic and real 3-D data.