Object reconstruction using the cooperation of 3D segments and 3D facets information

Given a series of trinocular images of an object, we have developed a method for building 3D Facets and 3D segments model of the object. From each triplet, a partial description of the object, called 3D View, is extracted. From the set of all extracted 3D Facets, a strategy for guiding the object reconstruction, based on a statistical method is developed. A 3D Matching Builder computes matchings between the 3D Primitives of consecutive 3D Views. Guided by the strategy, a Superstructure gathers all the matching informations given by the 3D Matching Builder in a set of equivalence classes. For each equivalence class of Superstructure, a representative is derived. The 3D Primitives model is finally computed merging informations of 3D Facet representatives and 3D Segment representatives. This method implemented in Smalltalk80, has been applied on a series of stereoscopic real images triplets; some results are provided at the end of the paper