Keywords: geometrical invariance, computer vision, model based object recognition
Start Date: 1 January 93 / Status: finished / Duration: 42 months
[ participants / contact]
Invariance is a novel approach to several of the difficult problems in computer vision. It has already had a major success in the development of practical industrial vision systems. The working group expect impact on the outstanding problems that have retarded practical applications of vision in inspection, surveillance, security, agriculture and medical imaging.
The objective is to build on existing work on model based recognition using geometrical invariants and extend the techniques to more general classes for recognition, and to visual tasks involving control such as navigation and tracking.
European links: The work builds on ESPRIT projects VIVA (6448), REALISE (8878) and CUMULI (21914) and the ACTS project VANGUARD (AC074).
A public domain suite of software to support applications involving geometrical invariance has been defined and released. These applications include view invariant recognition of 3D objects, site modelling and change detection, and 3D reconstruction from uncalibrated images.
The targeted C++ software suite, a public domain set of software libraries developed by GE, US, has been further developed and installed at a number of European sites including INRIA, Leuven University and Oxford University. This software supports the use of invariance in computer vision applications (e.g. model based recognition, multiple view geometry). The software is compatible with the emerging Image Understanding Environment.
Within VIVA-NSF, two workshops have been organised on applications of invariance in computer vision (Porta Delgada, Azores, 1993) and on applications and software (Martha's Vineyard, USA, 1996). The proceedings of the first workshop have been published by Springer-Verlag in the LNCS series, Vol. 825, 1994.
Oxford OX1 3PJ, UK
Oxford University, UK
Katholieke Universiteit Leuven, B
Stanford University, USA
University of Iowa, USA
University of Maryland, USA
Cornell University, USA
Dr. Andrew Peter Zisserman
Phone: +44 865 273 148
Fax: +44 865 273 908
VIVA-NSF - EC-US021, May 1997
please address enquiries to the ESPRIT Information Desk
html version of synopsis by Nick Cook