Attention dynamics in multiple object tracking

We present a computational model of multiple object tracking that makes trial-level predictions about the allocation of visual attention and the resulting performance. This model follows the intuition of allocated resources modulating spatial resolution, but it implements it in a specific way that leads to accurate predictions in multiple task manipulations. Experiments on human subjects, guided by the model’s predictions, demonstrate that observers tracking multiple objects use lowlevel computations of target confusability to adjust the spatial resolution at which the target needs to be tracked, and that the resulting allocation closely approximates the rational solution. Whereas earlier models of multiple object tracking have predicted the big picture relationship between stimulus complexity and response accuracy, our approach makes accurate predictions of both the aggregate effect of target number and velocity and of the variations in difficulty across individual trials and targets arising from the idiosyncratic within-trial interactions of targets and distractors.