| ECVision indexed and annotated bibliography of cognitive computer vision publications This bibliography was created by Hilary Buxton and Benoit Gaillard, University of Sussex, as part of ECVision Specific Action 8-1 The complete text version of this BibTeX file is available here: ECVision_bibliography.bib
I. A. Bachelder and A. N. Gove and M. Seibert and A. M. Waxman
ABSTRACT
We review progress on the development of a view-based neural system for unsupervised mapping and navigation within 3D environments defined by spatially distributed visual landmarks. The approach builds on our earlier work concerning unsupervised learning and recognition of 3D objects in the form of aspect categories and aspect transitions, effectively learning aspect graph representations. These neural systems are strongly motivated by physiological studies of cells in monkey inferotemporal cortex (cf. object recognition) and rat hippocampus (cf. environment mapping). The object learning system is summarized and illustrated for the case of aircraft objects in the visual domain. (It has also been applied to tactical targets in the SAR domain.) We compare our results to human psychophysical experiments on object inspection by Perrett. We then describe our environment learning system, which parcels the visual environment into a set of overlapping place regions. Architectures which learn both heading invariant and heading sensitive place categories are described, and results obtained on the mobile robot MAVIN are given. These results are compared to studies on rat hippocampus. Finally, we describe our approach to learning action consequences through the association of learned places and coarsely coded robot motions. |
|
| Site generated on Friday, 06 January 2006
|