SAVANT: A new efficient approach to generating the visual hull

Abstract

We present a new approach to generating a boundary representation (a polygonal mesh) of the visual hull from a set of silhouettes of an object taken from known camera positions. The approach uses spatial subdivision to compute the positions of vertices and then traverses the vertices around planar faces to produce a facetted representation of the visual hull. We show that, unlike standard approaches, the method is practical even for a large number of complex silhouettes. It is applicable to any 3D modeling system that uses the “shape-from-silhouette” approach to generating 3D models of objects. The approach can also be extended to efficiently compute polygonal mesh representations of the intersection of a set of arbitrary polyhedral models. This extends its applicability to CAD systems.