Informatisches Kolloquium Sommersemester 2011

Montag, 6. Juni 2011
um 17 Uhr c.t.
Vogt-Kölln-Straße 30
Konrad-Zuse-Hörsaal
Gebäude B

Early Cognitive Vision: Vision for Cognition (with Focus on Object and Action Grounding)

Prof. Norbert Krüger, PhD
Cognitive Vision Lab
The Maersk Mc-Kinney Moller Institute
University of Southern Denmark
Odense

We present an 'Early cognitive Vision' (ECV) system which is in particular designed as a front-end to allow for the realisation of cognitive processes to be build upon it, in particular in robotics. For this, we have been designing a visual representation which has a number of properties:

Local symbolic representation: Different local image structures become distinguished and represented by different symbolic descriptors in 2D and 3D. Richness and Orthogonality: The descriptors represent different aspects of visual information (in particular geometric and appearance information) in a way that allows for addressing this information in isolation. Structural Relations: Local information is embedded into spatial temporal relations expressing relevant structural knowledge. Hierarchy: Based on the structural relations the local entities are embedded into spatially and temporally extended entities. Disambiguation: The structural relations give rise to disambiguation processes that stabilize locally extracted and hence ambiguous information. Multi-purpose: The representation provides information to be used for multiple tasks and hence itself is task independent.

Basic design choices of the system have been motivated by the human visual system. The focus however is on an efficient machine vision system and not on a detailed analogy in the modelling of human visual processes. Our ECV system is available as a software library (www.covig.org) to be used as a front end by the scientific community focussing on the modelling of cognitive tasks.

We present the current status of the early cognitive system as well as a number of applications, in particular for object and action grounding and industrial robotics.

Kontakt

Prof. Dr. Jianwei Zhang

Telefon 2431