Purchase this article with an account.
Ashley M. Sherman, Tomás F. Yago Vicente, Gregory J. Zelinsky; Replacing the spotlight with a Kalman filter: A prediction error model of multiple object tracking. Journal of Vision 2014;14(10):358. doi: https://doi.org/10.1167/14.10.358.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Does multiple object tracking (MOT) rely on the predicted motion of an object? We addressed this using a continuous tracking paradigm (Exp 1), in which observers tracked 4 of 10 dots moving at 8°/s for 8s. On each 16.6ms frame, all dots had a .025 probability of turning 0°, 15°, 30°, 45°, 60°, 75°, or 90° in either direction (held constant within a trial, but varied across trials). Performance was best in the 0° condition, and declined as turn angle increased (p =.003). We used Kalman filters to model this effect of motion predictability on tracking by estimating the distance between a target's predicted and actual location, creating a "prediction field". Assuming an independent filter attached to each target, and that the probability of a swapping error increases with the number of times a distractor passes into a target's prediction field, we computed a swap potential for each target and averaged these to obtain a mean swap potential. We found that swap potential correlated highly with performance in Exp 1 (r = -.89, p <.01). We then tested our model on two other factors known to affect MOT performance: item speed (Exp 2), and trial duration (Exp 3). In Exp 2, speed ranged from 5-11°/s (p <.001). In Exp 3, speed was held constant at 8°/s, and duration ranged from 4-12s (p <.001). Both Exps 2 and 3 used a turn angle of 0°. We again found a high correlation between performance and swap potential (Exp 2: r = -.96, p =.001; Exp 3: r = -.95, p =.001). Taken together, these results not only suggest that observers are sensitive to the predictability of an item's motion, but also that a simple model of swapping potential built from prediction errors can estimate relative performance across a number of tracking tasks.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only