Multi-View Integration for a Rotating 3D Object

Abstract The aim of this paper is to propose a novel objectrepresentation for a 3D rigid object in tracking applications.A 3D volumetric object is bounded by a closed surface (2Dmanifold) on which visible features are perceived (e.g. color,texture). In our representation, we collect the topologicalstructures from the visible surface of the target object. As theobject rotates, new visible parts of surface would reveal newtopological structures. This collection of small 3D surfacepatches is reconstructed by processing topological completion.And the object representation is incrementally updatedfrom observing the target object in a video sequence.