A Computational Model for Visual Learning of Affordance-like Cues
Publikation aus Digital
Paletta L., Fritz G., Rome E., Dorffner G.
Proc. 29th European Conference on Visual Perception, ECVP 2006, St. Petersburg, Russia, August 20-25, Perception 35 , 2007
There are human affordances that are explicitly innate through evolutionary
development and there are affordances that have to be learned (Gibson
1979, Edwards et al., 2003, Brain Cognition, 495 - 502). In technical
vision systems, affordance based visual object representations are
function based and were so far predetermined by heuristic engineering
(Stark and Bowyer, 1995, Image Understanding 59 1 - 21). In contrast,
we propose that the selection of relevant predictive visual cues
should be performed by machine learning methodology, operating on
the basis of a complete spectrum of perceptual entities. In particular,
we investigate local gradient patterns in 2D (SIFT features, Lowe,
2004 International Journal of Computer Vision 60 91 – 110) in affordance
cueing among other visual modalities, such as colour, shape, and
3D information. Predictive features are then derived from attribute
based rules that are extracted from a decision tree based classifier
(Quinlan 1993). Decision trees are demonstrated to be capable to
provide a predictive feature configuration for the representation
of an affordance-like cue on the basis of an information theoretic
framework. Experimental results successfully verify the conceptual
framework from the viewpoint of an autonomous mobile agent that is
engaged within a robotic system scenario.
Keywords: affordances, robot vision, learning of cueing
Url: http://conferencesoft.com/ecvp/Program/Abstract.aspx?ag=282296c0-808d-4d6a-84fb-9263a3c578d6