A. Saminathan, A. J. Stoddart, A. Hilton, J. Illingworth
Centre for Vision, Speech and Signal Processing
University of Surrey, Guildford, Surrey GU2 5XH, UK
Most deformable surfaces used in computer vision are restricted to a simple underlying mesh topology. This is a considerable limitation when dealing with complicated objects whose topology is not known in advance. In previous work we introduced a deformable surface that can take on arbitrary topology while maintaining geometric continuity throughout.
In this paper we present some recent advances we have made that enable faster computation and a principled evaluation of the data fitting term. The method is used to obtain more compact representations for dense polygonal meshes obtained from multiple range images. We show results using a new approach to initialise the mesh topology.