Automatic registration of multiple range images based on cycle space

Abstract

This paper proposes a method for integrating multiple range images obtained by a range scanner, aiming to get an entire model automatically. We extract texture images for all range images and register any pair of them based on image features of their texture images. To verify the correctness and filter out mismatching pairs of range images, multiview matching process searches a model graph for all the consistent cycles. We prove that the consistent cycle space is a linear subspace and the entire cycle space can be decomposed into the direct sum of a consistent cycle space and its orthogonal complementary subspace. Moreover, we present an efficient algorithm for computing a basis of the consistent cycle space based on adjacency-of-cycle-basis graph, which avoids the exponential complexity of exhaustive search.