DESIGN OF A DESCRIPTOR FOR ALIGNING SURFACE RANGE IMAGES WITH APPLICATION TO ROBOTICS

Gerardo A. I. Pizo¹; José Maurício S. T. Motta¹

¹ Universidade de Brasilia

doi:10.20906/CPS/COB-2015-1619

Resumo

The 3D reconstruction of objects from range images comprises three steps: a) acquisition, in which the object surface is scanned from various viewpoints, generating a set of 𝑛 range images; b) registration, in which a rigid transformation M must be estimated to position images, peer to peer, in a single frame and c) matching, in which redundancies between registered views are treated to build a single and complete 3D model. This work proposes a methodology to register range images based on segmentation/reconstruction making use of an edge detection technique combined with a clustering technique using mesh decimation. The descriptor is applied to contours and it is invariant to similarity transformations including rotation, translation and uniform scale and also robust to noise. The proposed descriptor makes use of n corresponding points extracted from two images and a signature label is assigned specifically to a point considering the geometrical distribution of its neighborhood, facilitating the identification of possible areas of overlapping and reducing the ambiguity in the search process. The descriptor was evaluated through a series of tests with various object range images. To verify the validity of the candidate transformations, each of which passes through a fine alignment and the fitting errors between the two range images are evaluated by the ICP algorithm. The transformation that aligns the largest number of points is considered the solution.

Palavras-chave: Image Registration; Image Descriptor; ISO GUM; Range Imaging; Image Descriptors