Using simplified meshes for crude registration of two partially overlapping range images

Abstract

We present an efficient algorithm for crude registration of two partially overlapping range images. The central focus of our work is how to filter out the wrong pairs efficiently in order to reduce the search space. Motivated by the availability of robust and efficient mesh simplification algorithms, such as the QSLIM software package, which preserve the main characteristics of the surface shape, we propose to find a good initial transformation estimate on the basis of the vertices of simplified meshes, instead of the original dense sets of range data. Promising results, that make the iterative closest point matching procedure converges to a correct final alignment, are provided.