An automatic registration algorithm for the scattered point clouds based on the curvature feature

Object modeling by the registration of multiple range images has important applications in reverse engineering and computer vision. In order to register multi-view scattered point clouds, a novel curvature-based automatic registration algorithm is proposed in this paper, which can solve the registration problem with partial overlapping point clouds. For two sets of scattered point clouds, the curvature of each point is estimated by using the quadratic surface fitting method. The feature points that have the maximum local curvature variations are then extracted. The initial matching points are acquired by computing the Hausdorff distance of curvature, and then the circumference shape feature of the local surface is used to obtain the accurate matching points from the initial matching points. Finally, the rotation and translation matrix are estimated by the quaternion, and an iterative algorithm is used to improve the registration accuracy. Experimental results show that the algorithm is effective. ► A novel curvature-based automatic registration algorithm has been presented. ► The method can automatically solve the partially-overlapping 3D registration problem. ► The initial matching points are acquired by computing the Hausdorff distance of curvature. ► The circumference shape feature of the local surface is used to obtain accurate matching points.