In this paper, Bayesian Belief Networks (BBNs) technology is investigated in the light of a classical Computer Vision problem: that of inducing 3D world configurations from knowledge acquired by a multiple camera set. While the task can have, as indeed has, an evident practical interest, this paper does not aim at proposing BBN-based solutions to it. Rather, by illustrating how the problem could be formulated in the BBNs framework, we aim at highlighting potentialities and limitations of the approach. Three basic features will be empirically addressed in particular. flexibility, in the two-fold meaning of adaptability to different kinds of queries and capacity of incorporating available a-priori knowledge; accuracy, that is reliability of outcomes in presence of incomplete or noisy data; computational needs, that is CPU/memory demands and scalability issues
Models for BBM-Based Inference of Visual Properties
Caprile, Bruno Giovanni;Cattoni, Roldano;
1999-01-01
Abstract
In this paper, Bayesian Belief Networks (BBNs) technology is investigated in the light of a classical Computer Vision problem: that of inducing 3D world configurations from knowledge acquired by a multiple camera set. While the task can have, as indeed has, an evident practical interest, this paper does not aim at proposing BBN-based solutions to it. Rather, by illustrating how the problem could be formulated in the BBNs framework, we aim at highlighting potentialities and limitations of the approach. Three basic features will be empirically addressed in particular. flexibility, in the two-fold meaning of adaptability to different kinds of queries and capacity of incorporating available a-priori knowledge; accuracy, that is reliability of outcomes in presence of incomplete or noisy data; computational needs, that is CPU/memory demands and scalability issuesI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
