Object Recognition in Range Images Using CAD Databases

摘要

An aspect graph plays an important role in three dimensional object recognition.Its represents the three-dimensional shape of an object by its two dimensional qualitative views as seen from various viewpoints. To create the aspect graph of an object, the viewpoint space is partitioned into regions, each of which corresponds to qualitatively similar projections of the object. Algorithms for creating aspect graphs of polyhedral objects have been developed. We developed an algorithm to compute the aspect graph of a curved object. Our approach partitions the viewpoint space by computing boundary viewpoints from the shape descriptions of the object given in a computer aided design database. These computations are formulated from the understanding of visual events and the locations of corresponding viewpoints. We also studied new visual events for piecewise smooth objects.