Abstract
A novel algorithm is proposed for registration of two partially overlapping range images that are arbitrarily oriented. The algorithm is inspired by the Shape Index (SI) description of 3D geometry and the Scale Invariant Feature Transform (SIFT) algorithm for detecting and matching features between intensity images. For a range image pair to be registered, we first construct an artificial intensity image pair based on SI. Then, the SIFT algorithm is used to extract and match features between the artificial images, and accordingly the 3D feature matches between the range images can be established with ease. A RANSAC procedure, combined with a hysteresis thresholding scheme, is carefully designed to filter out the false 3D feature matches to ensure the robustness. Finally, the transformation is computed according to all correct 3D feature pairs to register the two range images. Experiments and comparisons are included to verify the superior robustness and effectiveness of the proposed approach.
