EUROGRAPHICS 2005 / J. Marks and M. Alexa (Guest Editors) Volume 24 (2005), Number 3 Exploiting the scanning sequence for automatic registration of large sets of range maps Paolo Pingi 1 , Andrea Fasano 2 , Paolo Cignoni 2 , Claudio Montani 2 , Roberto Scopigno 2 1 INOA - Istituto Nazionale di Ottica Applicata, Firenze, Italy 2 Visual Computing Lab, ISTI-CNR, Pisa, Italy Abstract Range map registration is still the most time consuming phase in the processing of 3D scanning data. This is because real scanning sets are composed of hundreds of range maps and their registration is still partially manual. We propose a new method to manage complex scan sets acquired by following a regular scanner pose pattern. Our goal is to define an initial adjacency graph by coarsely aligning couples of range maps that we know are partially overlapping thanks to the adopted scanning strategy. For a pair of partially overlapping range maps, our iterative solution locates pairs of correspondent vertices through the computation of a regular n × n kernel which takes into account vertex normals and is defined in the 2D space of the range map (represented in implicit 2D format rather than as a triangle mesh in 3D space). The shape-characterization kernel and the metrics defined give a sufficiently accurate shape matching, which has been proven to fit well the requirements of automatic registration. This initial set of adjacency arcs can then be augmented by the automatic identification of the other significant arcs, by adopting a criterion based on approximate range map overlap computation. With respect to the solutions present in literature, the simplifications and assumptions adopted make our solution specifically oriented to complex 3D scanning campaigns (hundreds of range maps). The proposed method can coarsely register range maps in parallel with the acquisition activity and this is a valuable help in assessing on site the completeness of the sampling of large objects. Categories and Subject Descriptors (according to ACM CCS): I.4.8 [Scene Analysis]: Range data Keywords: 3D scanning, automatic registration, coarse registration, range map alignment. 1. Introduction The increasing use of 3D scanning devices and the design of new and efficient algorithms for range data post-processing are the basis of a process where standard CAD tools are go- ing to be replaced by a semi-automatic process based on the direct sampling of real objects’ shape. Moreover, automatic acquisition of shape and appearance is no longer confined to the classical industrial applications (reverse engineering or quality control), but it is positively affecting new and impor- tant fields, such as Cultural Heritage (CH). Unfortunately, the creation of 3D digital models from reality is still far from being as simple as photography. The user has to man- age many complex processing steps (range maps acquisition, registration, fusion, geometry simplification, color attribute recovery). A current goal is to design new solutions which transform the scanning pipeline into an unattended process. In this direction, at present, it is possible to assert that the bottleneck of the whole process is the range maps registra- tion phase, since this is the only task where a considerable human intervention is still required. The accurate acquisition of a real object requires many range maps taken from differ- ent locations. If the scanner location and orientation are not tracked, all those range maps are produced in different co- ordinate spaces (each one depending on the corresponding unknown location and orientation of the scanner). The goal of the range map registration phase is to determine the rigid geometric transformations necessary to bring back all the co- ordinates of the acquired data into a unique Cartesian space. Registration is the fundamental precondition to merge all the data into a single and complete digital model. Our goal is to design new solutions to make the alignment of range maps an automatic process. c  The Eurographics Association and Blackwell Publishing 2005. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.