September 2015
Volume 15, Issue 12
Free
Vision Sciences Society Annual Meeting Abstract  |   September 2015
Simulating multiple object tracking performance using a Kalman filter model
Author Affiliations
  • Gregory Zelinsky
    Department of Psychology, Stony Brook University Department of Computer Science, Stony Brook University
  • Ashley Sherman
    Department of Psychology, Stony Brook University
  • Tomás Yago
    Department of Computer Science, Stony Brook University
Journal of Vision September 2015, Vol.15, 465. doi:10.1167/15.12.465
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Gregory Zelinsky, Ashley Sherman, Tomás Yago; Simulating multiple object tracking performance using a Kalman filter model. Journal of Vision 2015;15(12):465. doi: 10.1167/15.12.465.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Although previous work has debated whether motion extrapolation is used in multiple object tracking (MOT), here we show that a Kalman filter-based model of motion prediction can simulate tracking accuracy on a trial-by-trial basis. Behavioral data were collected in two experiments manipulating trial duration (exp 1; 4-12s at 8°/s) and item speed (exp 2; 5-11°/s at 8s). In each experiment observers tracked 4 of 10 moving discs. Consistent with previous findings, tracking performance was found to vary with both speed and duration (both p < .001). Our model attaches an independent Kalman filter to each target disc at the start of a motion sequence. At each time step of the simulation, each filter predicts the position and velocity of its associated disc, then decides new assignments of filters to discs probabilistically based on predicted positions and velocities and all observed disc positions. We define a 2D Gaussian distribution per filter, with the mean equal to the predicted position, and a covariance matrix that maximizes the probability-density-function in the direction of the predicted velocity. A disc is attached to the filter with a probability based on the pdf evaluated at the disc’s position, with tracking errors resulting when a filter becomes incorrectly attached to a non-target. The disks associated with filters at the end of each simulation are considered the items selected as targets. Accuracy is defined as the percentage of correctly selected targets. The model captured the behavioral decrease in tracking accuracy with both increasing trial duration (exp 1; p < .001) and increasing disc speed (exp 2; p < .001), although predicted accuracy in exp 2 was slightly higher than what was found in the behavioral data (p = .05). We conclude that motion prediction, simulated here by simple Kalman filters, can describe tracking accuracy in two widely-studied MOT tasks.

Meeting abstract presented at VSS 2015

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×