The Importance of Reasoning about Occlusions during Hypothesis Verification in Object Recognition

Charlie Rothwell 1
1 ROBOTVIS - Computer Vision and Robotics
CRISAM - Inria Sophia Antipolis - Méditerranée
Abstract : In this paper we study the limitations of current verification strategies in object recognition and suggest how they may be enhanced. On the whole {\em object topology} is exploited little during verification. In practice, understanding the connectivity relationships between features in the image, or on the object, can lead to significantly more accurate evaluations of recognition hypotheses. Usually adjacent features on an object should be party to mutual visibility constraints. This is to say that when a model is hypothesized in a scene, two features which are adjacent within the model of an object should either both be visible in an image of the object, or both occluded. If not, then we can broadly say that an occlusion event should exist between the two features. In the case that this event fails to be measurable, we can start to infer that the model hypothesis is incorrect. Similar reasoning can be used to exploit image topology and the uniqueness of sets of model-image correspondences. Generally, such a line of thinking departs from traditional approaches in which topological interactions between features are not exploited fully. Testing out our algorithms for topology and occlusion analysis has involved the implementation of a complete object recognition system. The system we have built measures planar algebraic invariants in real images and uses these to index into a model base. The result of the indexing step is a list of hypotheses. These hypotheses are evaluated both using traditional verification algorithms, and also using our more detailed methods.
Type de document :
Rapport
RR-2673, INRIA. 1995
Liste complète des métadonnées

https://hal.inria.fr/inria-00074017
Contributeur : Rapport de Recherche Inria <>
Soumis le : mercredi 24 mai 2006 - 14:18:20
Dernière modification le : samedi 27 janvier 2018 - 01:31:27
Document(s) archivé(s) le : jeudi 24 mars 2011 - 13:50:22

Fichiers

Identifiants

  • HAL Id : inria-00074017, version 1

Collections

Citation

Charlie Rothwell. The Importance of Reasoning about Occlusions during Hypothesis Verification in Object Recognition. RR-2673, INRIA. 1995. 〈inria-00074017〉

Partager

Métriques

Consultations de la notice

125

Téléchargements de fichiers

171