Solid Modelling Data Structures For Computer Vision

The data structures and access mechanisms used in representing objects in computer vision have generally been of little interest. While a conceptual description of the information in a symbolic scene representation is frequently described, the actual underlying structures for this representation are normally ignored. As more parallel approaches are applied to the scene Model generation task, this becomes less adequate. In an environment typified by multiple, competing models being modified by several processes, it is necessary to pay more attention to the question of maintaining a consistent model structure. In addition, the careful development of a general purpose scene modelling environment provides the ability to pay more attention to issues in the vision task, and less to symbolic representational issues. This paper describes a solid modelling environment for computer vision. The modelling system implements a hierarchical object structure, with points, lines, polygons, rigid objects, and articulated objects. User programs are able to freely manipulate these entities, without concern for the underlying representations. Model manipulation is defined in terms of graph grammar productions.