The use of characteristic views as a basis for recognition of three-dimensional objects
Indranil Chakravarty · 1982
This thesis describes a new technique for the representation of three-dimensional objects that is applicable to automation tasks such as recognition and assembly. The primary objective of this representation scheme is to reduce a 3D model into a canonic 2D model that can be used to determine object identity, orientation and location in 3D space. The reduction in dimensionality is based on the notion of factoring the space of all possible perspective projections of an object into a set of characteristic views, where each such view defines a family of projections which are topologically identical. Each characteristic view may be associated with one or more stable positions of the object. A linear transformation is then developed that allows mapping some member of the set of characteristic views to any arbitrary projection of the object to determine orientation and location of the object in 3D space. This transformation is used both for identification and for determining the stable position of the CV into which the arbitrary projection can be transformed. A line and junction labeling scheme is developed that is used for classifying the characteristic views into a hierarchy, based on the degeneracy of the projection. The labeling constraints are used at a later stage for guiding the search for matching two line structures. The problem of matching and correlating points between the characteristic view and an arbitrary projection is accomplished in two stages. In the first stage the gray-level image is reduced into a binary image describing the object line-structure. The line-structure is chain-encoded, labeled and the junctions classified. In the second stage a stage-space technique is used for matching silhouettes of the extracted line-structure to the stored characteristic-views. The semantics associated with the lines and junctions are used to prune the search space. Once the silhouettes are matched, one proceeds to determine the transformation that best maps the projection to the stored characteristic view. The transformation is then used for verifying other junction and line labels in the extracted line structure.