Abstract
We hypothesize that our ability to track objects through occlusions (and near occlusion events; NOEs) is mediated by timely assistance from gaze in the form of “rescue saccades”–eye movements to tracked objects that are in danger of being lost due to impending occlusion. Subjects tracked 2-4 target sharks (out of 9) for 20 seconds as they swam through a rendered 3D underwater scene. One of the sharks was then probed, and the task was to indicate whether the probed shark was one of the targets. Tracking accuracy with 2-3 targets was >95%, but dropped to 74% with four targets. This tracking performance was mirrored in the frequency of rescue saccades. For trials in which the probed object was correctly identified as a target, the probabilities of a saccade accompanying an occlusion were .52 and .51 in the Track 2 and 3 conditions, but only .34 in the Track 4 condition (p<.001). This low Track 4 frequency of rescue saccades did not differ from the rescue saccade rate on incorrect Track 4 trials (.29), although rescue saccades in all target conditions occurred far more frequently than saccades to targets during non-occlusion tracking (≤.18). Analyses of the distances between targets and the nearest other object also revealed the expected increase in occlusions and NOEs with the number of targets (creating more opportunities for object confusions), as well as an inverse correlation between rescue saccades and this nearest-object distance. We interpret these data as evidence for a tracking system that monitors tracked objects for events that might result in track loss (e.g., occlusions and NOEs), and requests help from the oculomotor system to resolve these momentary crises. Because rescue saccades are inherently serial, as the number of these crises increase with targets, some requests for help go unsatisfied, resulting in degraded tracking.
This work was supported by NIH grant RO1-MH063748 and NSF grant IIS-0527585.