The use of characteristic view classes for three-dimensional object recognition
Ruye Wang · 1991
A characteristic-view (CV) class of a planar-faced solid (PFS) is a class of representative visible-edge projections of the solid such that all members of the class are topologically equivalent, have the same planar junction graph, and can be related by a geometric transformation. The viewing space of a PFS can be partitioned into regions such that a view of the PFS from any location within a region belongs to the same CV class and is different from all the CVs of the neighboring regions. These regions are called characteristic-view domains (CVD). This thesis presents a method for partitioning the viewing space into CVDs and to find the complete set of CV classes for a given PFS. Also, it presents some new results relating to the second view problem (see Duda and Hart's book Pattern Recognition and Scene Analysis). Based on these new developments, the thesis describes a new method for model-based object recognition. Algorithms for representing, sorting, classifying and geometrically and graphically matching CVs are also presented. All of these operations are utilized in the proposed object recognition method. Experimental results are included for both synthesized and real objects.