Which way did it go? Measuring trajectory information in multiple object tracking

Skyler S Place 1, Todd S Horowitz 1,2

1 - Brigham and Women's Hospital
2 - Harvard Medical School

In multiple object tracking (MOT) experiments, observers are capable of tracking 3-5 targets among identical distractors. What information about target trajectories can the visual system use in this task? We present a new method, which allows us to measure the precision of observers’ information about object motion direction in MOT.

In Exp. 1, we measured directional information for targets. Observers tracked 4 of 8 moving disks for 5 seconds. At the end of the trial, a green arrow appeared on one target disk. The arrow was presented at an angle (±15°, ±30°, ±±45°, or ±90°) to the target’s actual direction of motion. The task was to make an unspeeded response indicating whether the green arrow was rotated clockwise or counterclockwise relative to the true direction. The 70% accuracy threshold was approximately 30°.

In Exp. 2, we compared directional accuracy for targets and non-targets. Since the directional task probed non-targets as well as targets, we introduced two incentives for observers to track targets. First, on 20% of trials, a single disk turned blue, and observers made an unspeeded 2AFC decision as to whether the blue disk was a target; directional data were collected on the remaining trials. Second, we introduced more non-targets than targets (observers tracked 3 of 8). Accuracy for target direction was 86%, while non-target responses were near chance.

The visual system has access to coarse information about target trajectories in MOT. This information is only available for attended objects.