Gaze Guidance


Gaze Guidance is a growing area in the field of Human-Computer Interaction. The goal is to support the viewer during visual inspection of his/her environment by giving suggestions of where to look (Barth et al 2004). Gaze guidance is potentially applicable in situations where the viewer is confronted with a large visual display (or visual field), which needs to be searched for specific information (PC monitor, car cockpit, medical image analysis, Virtual Reality, gaming).


[link to project website]    the website of the 5-lab European collaboration

[pdf manuscript]              manuscript full of ideas for human-computer interaction

[talk]                                slides about this issue


My task is to find gaze-capturing events, with which one can attract a viewer’s gaze to preferred locations in a ‘non-irritating’, comfortable manner. The following is a summary of aspects, which one has to consider when building such a system:




This would be the situation for a car cockpit:



But when approaching such a system it may also be useful to firstly implement and test simpler guidance tasks. In my manuscript (pdf), I provide a number of ideas for gaze guidance on a PC monitor. One specific idea I pursue is the Gaze-Recapturing Editor Cursor together with Michael Dorr.


Method: Currently we are testing the idea of gaze guidance using a letter detection and identification search using the following type of display (it is one frame of a dynamic-noise movie, letters are shown schematically only).


There exists saccadic undershoot…


and this is the summary graph:




Summary of recommendations for designing gaze guidance markers:


1) Aspect range: To compensate for the decline in peripheral acuity, the marker’s amplitude is increased with eccentricity by an exponentially saturating function: amrk(e) = amin + amax-exp(-e)amax (amin = minimal amplitude, amax = maximal amplitude).


2) Aspect location: If a compensation for undershoot is desired, the marker should be placed radially beyond its target by 18% of target eccentricity. Such compensation is probably required when small, hard-to-detect targets are to be foveated which are embedded in a complex background.


3) Aspect appearance: a) Motion markers are better gaze-capturing events than stationary markers, however they are potentially detrimental to recognition performance at their location.

b) If one uses a luminance marker, which is merely added to the luminance profile to make it just-noticeable, it may be necessary to set a lower bound in order to avoid the ‘neglect’ of very low luminance markers.


4) Aspect occurrence: In case of guidance toward briefly appearing stimuli, the optimal gap size between marker offset and target onset is ca. 100ms to avoid strong forward-masking effects.