Purchase this article with an account.
Gregory Zelinsky, Mark Neider, Andrei Todor; Multi-object tracking in a realistic dR environment. Journal of Vision 2007;7(9):895. doi: 10.1167/7.9.895.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Multi-object tracking (MOT) has traditionally been studied using simple patterns moving in two dimensions. We explored MOT using more realistic objects moving in three dimensions. Our stimuli were 9 sharks in a rendered 3D underwater scene. The sharks started each trial arranged in a ring, with 1–4 marked for tracking. All of the sharks then swam throughout the scene for 20 seconds, after which one of the sharks was probed. The task was to indicate whether the probed shark was one of the marked sharks. We manipulated whether the probed shark intersected another shark during its movement, and whether it appeared near or far in depth at the time of probe onset. Consistent with the 2D MOT literature, accuracy declined non-linearly with the number of objects to track. Tracking was near perfect ([[gt]]95%) with 1–2 sharks, still good with 3 sharks (87%), and quite poor with 4 sharks (62%). Tracking was relatively unaffected by intersecting trajectories or depth of the probed shark on the final frame. Fixation analyses revealed that the proportion of gaze and the maximum dwell time on any one shark decreased as the number of tracked sharks increased. This suggests that observers tended to look back-and-forth between sharks in the track-two condition, but not when tracking more objects, and that gaze was not being held on any one shark when multiple sharks were being tracked. Instead, observers appear to have adopted a centroid tracking strategy, choosing to fixate an average location from which to peripherally track multiple sharks. We conclude that tracking benefits from the periodic fixation of objects, which may serve to maintain track on an object during moments of potential confusion. As the number of tracked objects increase and observers shift to a centroid tracking mode, such “foveal rescues” become infrequent and tracking performance suffers.
This PDF is available to Subscribers Only