Combining meshes and geometric primitives for accurate and semantic modeling

We propose an original hybrid modeling process that represents three-dimensional (3D) models as a combination of meshes and 3D-primitives. Meshes describe details such as ornaments or statues, whereas 3D-primitives code for regular shapes such as walls or columns. Starting from a mesh obtained by multi-view stereo techniques, these primitives are substituted to the mesh where they are detected. This strategy allows the introduction of semantic knowledge, the simplification of the modeling, and even corrections of errors generated by the acquisition process. We design a two step method consisting in first, segmenting the mesh using a multi-label energy model optimized by classical techniques, and then, fitting 3D-primitives such as planes, cylinders or tori on the obtained partition where it is relevant. Experiments on both real meshes and synthetic data show the good potential of the proposed approach.