ressearch project afm

parlab

research> activities

Research projects

Attention in 3D Vision

Nabil Ouerhani & Tim Jost

Project program

SNFS, project no. 2153-064894

Partner: WOF, Neurologische Klinik, Inselspital Bern, PD. Dr. R. Mueri

Keywords

Computer vision, human visual system, visual attention, eye movements, depth perception, 3D imaging, three-dimensional vision

Purpose of the project

The objective of this project is to study the contribution of depth in the mechanism of visual attention and to supplement the existing models of visual attention by a detailed consideration of this new component. The expected benefit is an improved understanding of attention mechanisms relative to natural 3D vision and the useful application of 3D attention in computer vision.

This objective is significant and timely for several reasons.

• Until recently, the depth did not constitute significance for computer vision simply by the fact that this channel of information was technically lacking, but today, when 3D-cameras that deliver the scene depth become available, adequate use of this source of information is expected.

• Also, in human vision, the mechanism and the role of visual attention related to depth are not studied in detail.

• Eye movements are intimately linked to vision and visual attention, and it is generally accepted that eye movements and the concomitant focal visual processing have advantages for human and machine vision. Therefore, a methodological link between eye movement analysis of a visual scene in humans and the modeling of computer vision seems to be promising.

• The recent availability of autostereoscopic screens constitutes a new key to extend experimentation sites and to allow now the practical investigations of eye movements related to depth changes.

• Finally, the interdisciplinary theory-empiricism cycle proposed in this project allows solving critical problems more quickly than when staying within the monodisciplinary approach.

By linking human eye movement data during visual attention experiments in 2D and 3D space, the proposed work will provide new ground for the modeling of a fundamental component of the human visual perception and new insights for computer vision methods.

First experiments

In 1999, first contacts between WOF and IMT were established, and preliminary experiments concerning computer and human exploration were performed (figure 1)

Figure 1: Comparison of human (left) and computer generated (right) visual attention. Qualitatively, there is a clear correlation between fixations (green dots) and attention spots calculated by the computer (source WOF & IMT)

This led us to recognize the potential of comparing human exploratory behavior and computer behavior as a means to investigate attention. We found that a systematic comparison of human exploratory behavior and computer behavior was not done so far and came therefore to the formulation of the present project

References

to be announced