Eye tracking data to measure the effect of target shape on smooth pursuit eye movement

ApparatusThe gaze data of participants was recorded using a wearable eye-tracking device (Pupil Labs Core device). This device takes the shape of a pair of goggles with three embedded cameras. Two cameras (192 x 192 px, 120FPS) are used to record videos of the eyes of the wearer while one wide angle lens world camera (1080p, 120FPS, FOV 139°x83°) records its field of view.

The device was connected via USB to a computer that was running the dedicated Pupil Capture software to perform calibration and validation of the gaze data and handle the recording of the experiments (video flux, gaze data, surface tracking).

Participants were seated 60 cm way from the screen on which the stimuli were displayed. Participants' head was constrained using a chin rest to minimize head movements during the experiment.

The screen was monitor with a resolution of 1920 x 1080 pixels and width of 47.60 cm.

StimuliWe used seven different shapes for the moving target, namely:
B: a plain black circleAB: a circle with a bull's eyeABC: a circle with a bull's eye and a cross hairABC_rotated: a circle with a bull's eye and a cross hair rotated such that one arm of the cross hair was aligned with the direction of motionconcentric: a set of concentric circlesdilating: a set of concentric circles that dilate, i.e. the target is animated and the concentric circles move away from the centercontracting: a set of concentric circles that contract, i.e. the target is animated and the concentric circles move towards the center
The targets have a size of 1.5°.

We follow a procedure similar to the one used by Liston and Stone [1]. We use the classic step-ramp paradigm [2]. The target is first displayed at the center of the screen, with a reduced opacity to signify that the system is waiting for the participant's input. When the participant presses a button, the target becomes fully opaque and the stimulus starts. The target is immobile for a randomized duration (drawn from an exponential distribution with a mean of 3 s, and truncated between 2 and 4 seconds). This serves the purpose of ensuring that the participant is not able to predict the start of the motion. The target is then offset from the center of the screen in a direction opposite to the direction of motion. It then follows a straight trajectory at a constant speed until it is 10° away from the center of the screen. The initial offset is computed such that the target will be at the center of the screen again 200 ms after the start of the motion. This is done to avoid the initial catch-up saccade that would occur if the target was already directly moved away from the center of the screen.

For each target shape, we use 24 different direction of motion (from 0 to 23 x pi/12 radians, in steps of pi/12 radians) and 5 different speeds (from 16 to 24°/s, in steps of 2°/s). This results in a total of 120 different stimuli per target. The stimuli are presented in a randomized order to the participants, in 6 blocks of 20 stimuli. Each block is followed by a break to avoid fatigue. The experiment was repeated on 2 different days, with a minimum of 1 days between the two sessions.

CalibrationThe device was adapted to the face morphology of each participant, in order to obtain a good view of both pupils. Then, each block started with a calibration procedure, using a custom 9 points calibration. These multiple calibration ensure that the tracking performance are not impacted by slippage of the headset, as could be the case if only one calibration was made at the start of the experiment.

We opted for the 2D calibration provided in Pupil Core's software because it is deemed more precise than the 3D calibration and is well-suited for gaze tracking on a simple surface, such as a monitor, with limited participant motion.

Although Pupil Core inherently provides gaze position coordinates in 3D space, the Pupil Player software can compute 2D gaze positions on a specified surface. To achieve this, the surface must be defined using April markers and registered in the software. The world camera detects these markers, allowing the software to map the gaze position onto the defined surface.

Data structureThe data is organized as follows:

{id_participant}
├── {target_shape}
│ ├── {block_id}
│ │ ├── annotations.csv
│ │ ├── gaze_positions.csv
│ │ ├── pupil_positions.csv
│ │ ├── blinks.csv
│ │ ├── gaze_positions_on_surface_experiment_adrien.csv
| ├── {block_id}
│ │ ...
├── {target_shape}
│ ...

annotations.csv: contains the annotations sent to the Pupil Capture software during the experiment. In particular, it contains the start and end timestamps of each stimulus presentation, as well as the direction of motion and speed of the target.gaze_positions.csv: contains the gaze positions of the participant in 3D space (see the Pupil Core documentation for more details).pupil_positions.csv: contains the pupil positions of the participant (see the Pupil Core documentation for more details).blink.csv: contains the timestamps of the blinks detected by the Pupil Player software.gaze_positions_on_surface_experiment_adrien.csv: contains the gaze positions of the participant on the surface defined in the Pupil Capture software.
References[1] Dorian B. Liston and Leland S. Stone. “Oculometric Assessment of Dynamic Visual Processing”. In: Journal of Vision 14.14 (Dec. 19, 2014), pp. 12–12.
[2] Cyril Rashbass. “The Relationship between Saccadic and Smooth Tracking Eye Movements”. In: The Journal of Physiology 159.2 (Dec. 1961), pp. 326–338