Shape from contours and multiple stereo - A hierarchical, mesh-based approach

We present a novel method for 3D shape recovery based on a combination of visual hull information and multi image stereo. We start from a coarse triangle mesh extracted from visual hull information. The mesh is then hierarchically refined and its vertex positions are optimized based on multi image stereo information. This optimization procedure utilizes 3D graphics hardware to evaluate the quality of vertex positions, and takes both color consistency, and occlusion effects as well as silhouette information into account. By directly working on a triangle mesh we are able to obtain more spatial coherence than algorithms based entirely on point information, such as voxel-based methods. This allows us to deal with objects that have very little structure in some places, as well as small specular patches.