Computer vision based automatic construction of 3-D geometry scene models

Describes an approach towards the automatic construction of 3-D object models using images of the scene. The method makes use of a visible surface scene representation called the 2/sup 1///sub 2/-D sketch. The implementation utilises multiple simulated 2/sup 1///sub 2/-D sketches to construct a 3-D voxel based geometric model. The depth information from each 2/sup 1///sub 2/-D sketch is used for chipping voxels from a 3-D array of voxels. This process is analogous to a sculptor chipping a stone block, using spatially displaced views, to create an imitation of the scene. The geometric model forms a basis for deriving a more complete and visually realistic 3-D graphic model of a scene.<
>