Work Area: Computer Vision
Keywords middle-level vision, optic flow, cue integration, surfaces, objects
Start Date: 1 August 92 / Duration: 36 months / Status: running
[ participants / contact ]
Abstract INSIGHT II aims at gaining understanding of the computational problems of natural vision and their solution in biological and silicon systems. The project draws on the disciplines of psychophysics, neurophysiology and computer vision to study key areas including the measurement of optic flow, cue integration, task-dependent use of optic flow, and surface and object representations. The work builds on results achieved in INSIGHT I (3001).
The aim of INSIGHT II is to improve our understanding of vision at the computational level, that is, to identify the representations to be used and the algorithms to compute them for general-purpose vision systems operating in a fluctuating natural environment.
Using the results of a previous action (INSIGHT, 3001) as a guide, a number of key areas have been identified, mainly in middle-level vision, where rapid progress can be expected. These areas include the measurement of optic flow, cue integration, the task-dependent use of optic flow, and surface and object representation. These five topics are approached from a multidisciplinary point of view combining psychophysics and neurophysiology in studying biological visual system in primates, and computer science in studying computer vision. While psychophysics studies human or primate performance and conjectures about underlying mechanisms, physiology measures these mechanisms directly. The strategies employed by the most successful computer (the primate brain) will thus be confronted with those used in computer vision to solve similar problems.
The discovery of the site of convergence of luminance, motion and texture cues specifying shape in the primate brain.
New insights into the receptive field structure of higher order visual neurons and on the integration of visual and eye movement signals in the primate brain.
On the psychophysical side, a new technique has been developed for the measurement of subjects' perception of surfaces. A wealth of new data on the estimation by humans of surface position, orientation and curvature specified by disparity, motion, luminance, texture and other cues has been obtained.
On the computational side, a new theory has been developed on the recovery of surfaces from the structure of the disparity field.
In computer vision, the consortium has been extremely active at all levels, mathematical theory, implementation and experimentation, with the estimation of optic flow and its use for recovering object 3D structure and 3D motion.
The vision knowledge gained from these fundamental studies will give firm ground to more applied projects in which vision is used to guide robots or recognise patterns.
The work of this consortium is presented at the leading conferences in computer vision, neuroscience and psychophysics.
The work of INSIGHT I has been published in Artificial and Biological Vision Systems, eds. G.A. Orban and H.-H. Nagel, Springer Verlag, 1992.
Katholieke Universiteit Leuven - B
Laboratorium voor Neuro- en Psychofysiologie
B - 3000 LEUVEN
Universitätsspital Zurich - CH
Ruhr Universität Bochum - D
Universität Karlsruhe - D
INRIA - F
Università di Genova - I
Rijksuniversiteit Utrecht - NL
Royal Institute of Technology Stockholm - S
University of Oxford, Dept. of Experimental Psychology - UK
University of Oxford, Dept. of Engineering Science - UK
University of Sheffield - UK
Katholieke Universiteit Leuven, ESAT - B
Université de Nice-Sophia Antipolis - F
Computer Technology Institute, Patras - GR
Stirling University - UK
Prof. G.A. Orban
INSIGHT II - 6019, August 1994
please address enquiries to the ESPRIT Information Desk
html version of synopsis by Nick Cook