by Marco Gavanelli
Università degli Studi di Ferrara
2002
Abstract
Artificial Vision is an important field of Artificial Intelligence: the human brain is able
to derive various types of information about the outer world from vision. However, the
semantic interpretation of objects is a hard task for a computer. Constraints have been
widely used for visual recognition, because they are able to describe very complex models
and situations in a natural way.
Constraint Logic Programming is a new paradigm of declarative programming languages that is able to deal declaratively and efficiently with problems described with constraints.
In this thesis, we addressed problems that stem from Artificial Vision with constraints.
We developed a 3D visual search system able to detect an object described with constraints
in an image. We provide constraint models for dealing with the problems in Artificial
Vision. We generalize the applied techniques to other important fields of Artificial Intel-
ligence, developing new tools of Constraint Satisfaction and Optimization in Constraint
Logic Programming. In order to obtain efficiency, we propose a tight interaction between
the Constraint Solver and the provider of visual features. We give a theoretical model,
solving algorithm and a language extension.
We model 3D objects with an object-centered model, based on visual information. From
each viewpoint, only part of the visual features of a 3D object can be seen; we model objects
by adding invisible features in the domains and select only the most informative assignments
by means of a partial order.