Abstract
The primary purpose of saccadic eye movements is to bring the image of visual objects of interest onto the fovea. However, this does not imply that all purposive saccades in a visual environment are visually guided, since visual crowding can render less salient objects in a visual scene invisible if they are located far from present gaze. Moreover, after repeatedly fixating such a non-salient element in a visual scene one senses that one's eye lands right on the target, despite its invisibility. We sought to confirm this phenomenological finding by recording saccades of 3 subjects using an S-R Eyelink video eyetracker (250 Hz) while they made saccades in a complex visual scene displayed on a computer monitor (∼40 × ∼30 deg). Prior to recordings, subjects freely viewed the display and were told the identity of possible saccade targets in the scene. In the invisible task, initial fixation position was varied trial-by-trial in a region from which all possible targets were invisible due to crowding and distance. Target was held constant within a block of 16 trials, and performance on the last 8/16 trials was analyzed. Saccade endpoint errors and peak velocities in the invisible task were compared with those in visually guided and memory-guided delayed tasks. Phenomenology was confirmed: metrics in the invisible task were comparable to the visually guided task, whereas memory-guided saccades had greater endpoint errors and lower peak velocities. Additional experiments demonstrated that up to 3 targets could be stored simultaneously in this memory system. These results suggest that a scene-based memory system enables the generation of saccades to invisible targets as if they actually were visible. Such a system could prove useful both for the continuous monitoring of non-salient objects in a crowded visual scene and for refixating more salient stimuli without having to tax peripheral visual processing.
Supported by: AFOSR 13586-43-4-6557, K08 22349082409