cnnmon . refmema ' ' In model based object recognition, the primary objective is to effrciently match features which have been extracted from sensory data to corresponding features in object models; this being done with the constraint that relations between the features in the object models are m'rrored by the relations between the features extracted from the sensory data. A problem which confronts this process is the dificulty in extracting features and relations from sensed data, and precisely determining the values of their relevant attributes. Furthermore, it is o fen the case that the features which are visible from a single viewpoint are not sufficient to uniquely identib the object and its pose. In each of these cases, a system is needed which can formulate and associate credibilities with hypotheses about the possible ia'entities and poses of the objects in the scene. This paper describes an architecture for reasoning with uncertainty about the identities of objects in a scene. The main components of this architecture create and assign credibility to object hypotheses based on feature match, object. relational , and aspect consistencies. We use the Dempster-Shafer formalism for representing uncertainty, so these credibilities are expressed as belieffunctions which are combined using Dempster's combination rule to yield the system's aggregate belief in each object hypothesis. One of the principal objections to the use of Dempster's rule is that its worst case time complexity is exponential in the size of the hypothesis set. We will show how the structure of the hypothesis sets developed by our system allow for a polynom'al time implementation of the combination rule.
Download Full PDF Version (Non-Commercial Use)