Limits...
Smaller is better: drift in gaze measurements due to pupil dynamics.

Drewes J, Zhu W, Hu Y, Hu X - PLoS ONE (2014)

Bottom Line: Recently, a significant impact of changes in pupil size on gaze position as measured by camera-based eye trackers has been reported.We observed a wide range of drift direction, mostly downward and nasal.We demonstrate two methods to partially compensate the pupil-based shift using separate calibrations in pupil-constricted and pupil-dilated conditions, and evaluate an improved method of compensation based on individual look-up-tables, achieving up to 74% of compensation.

View Article: PubMed Central - PubMed

Affiliation: Centre for Vision Research, York University, Toronto, Canada; Center for Mind/Brain Sciences, Trento University, Rovereto, Italy.

ABSTRACT
Camera-based eye trackers are the mainstay of eye movement research and countless practical applications of eye tracking. Recently, a significant impact of changes in pupil size on gaze position as measured by camera-based eye trackers has been reported. In an attempt to improve the understanding of the magnitude and population-wise distribution of the pupil-size dependent shift in reported gaze position, we present the first collection of binocular pupil drift measurements recorded from 39 subjects. The pupil-size dependent shift varied greatly between subjects (from 0.3 to 5.2 deg of deviation, mean 2.6 deg), but also between the eyes of individual subjects (0.1 to 3.0 deg difference, mean difference 1.0 deg). We observed a wide range of drift direction, mostly downward and nasal. We demonstrate two methods to partially compensate the pupil-based shift using separate calibrations in pupil-constricted and pupil-dilated conditions, and evaluate an improved method of compensation based on individual look-up-tables, achieving up to 74% of compensation.

Show MeSH
Drift compensation results.The top graph shows the profile of the background luminance time course (black dashed line) and the pupil size response (normalized, then averaged across subjects). The left group of bars shows the average rectified measurement deviation per 2-second interval on the same timescale (mean and 1 s.e.m.). The single right bar graph shows the average across the entire trial duration (10 s). Gray bars represent uncorrected measurements. Blue bars represent measurements compensated according to the 2-Point method. Green bars represent the 3-Point method. Red bars represent the LUT-Method.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4206464&req=5

pone-0111197-g005: Drift compensation results.The top graph shows the profile of the background luminance time course (black dashed line) and the pupil size response (normalized, then averaged across subjects). The left group of bars shows the average rectified measurement deviation per 2-second interval on the same timescale (mean and 1 s.e.m.). The single right bar graph shows the average across the entire trial duration (10 s). Gray bars represent uncorrected measurements. Blue bars represent measurements compensated according to the 2-Point method. Green bars represent the 3-Point method. Red bars represent the LUT-Method.

Mentions: All three attempted approaches succeeded in reducing drift magnitude, to varying amounts. Averaged time courses of compensated and raw data can be seen in Figure 4. Averaged across subjects and time, both the 2-Point and 3-Point methods reduced the amount of measured drift to 43% of the original value, while the LUT method was able to reduce the drift to only 26%. When dividing the fixation trials into the 5 basic periods (dark, luminance increase, bright, luminance decrease, and again dark, see Figure 5), it can be seen that of the three methods, the LUT method works best in the first four time intervals, while it achieves the worst compensation in the last time interval. This reveals the weakness of the look-up table approach: as the table was built from the constriction flank of the pupil size time course, the compensation is highly effective during the beginning of the trial, but less effective during the later, dilation-dominated phase. The 3-Point approach is actually just slightly superior to the 2-Point approach in all time intervals except the first.


Smaller is better: drift in gaze measurements due to pupil dynamics.

Drewes J, Zhu W, Hu Y, Hu X - PLoS ONE (2014)

Drift compensation results.The top graph shows the profile of the background luminance time course (black dashed line) and the pupil size response (normalized, then averaged across subjects). The left group of bars shows the average rectified measurement deviation per 2-second interval on the same timescale (mean and 1 s.e.m.). The single right bar graph shows the average across the entire trial duration (10 s). Gray bars represent uncorrected measurements. Blue bars represent measurements compensated according to the 2-Point method. Green bars represent the 3-Point method. Red bars represent the LUT-Method.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4206464&req=5

pone-0111197-g005: Drift compensation results.The top graph shows the profile of the background luminance time course (black dashed line) and the pupil size response (normalized, then averaged across subjects). The left group of bars shows the average rectified measurement deviation per 2-second interval on the same timescale (mean and 1 s.e.m.). The single right bar graph shows the average across the entire trial duration (10 s). Gray bars represent uncorrected measurements. Blue bars represent measurements compensated according to the 2-Point method. Green bars represent the 3-Point method. Red bars represent the LUT-Method.
Mentions: All three attempted approaches succeeded in reducing drift magnitude, to varying amounts. Averaged time courses of compensated and raw data can be seen in Figure 4. Averaged across subjects and time, both the 2-Point and 3-Point methods reduced the amount of measured drift to 43% of the original value, while the LUT method was able to reduce the drift to only 26%. When dividing the fixation trials into the 5 basic periods (dark, luminance increase, bright, luminance decrease, and again dark, see Figure 5), it can be seen that of the three methods, the LUT method works best in the first four time intervals, while it achieves the worst compensation in the last time interval. This reveals the weakness of the look-up table approach: as the table was built from the constriction flank of the pupil size time course, the compensation is highly effective during the beginning of the trial, but less effective during the later, dilation-dominated phase. The 3-Point approach is actually just slightly superior to the 2-Point approach in all time intervals except the first.

Bottom Line: Recently, a significant impact of changes in pupil size on gaze position as measured by camera-based eye trackers has been reported.We observed a wide range of drift direction, mostly downward and nasal.We demonstrate two methods to partially compensate the pupil-based shift using separate calibrations in pupil-constricted and pupil-dilated conditions, and evaluate an improved method of compensation based on individual look-up-tables, achieving up to 74% of compensation.

View Article: PubMed Central - PubMed

Affiliation: Centre for Vision Research, York University, Toronto, Canada; Center for Mind/Brain Sciences, Trento University, Rovereto, Italy.

ABSTRACT
Camera-based eye trackers are the mainstay of eye movement research and countless practical applications of eye tracking. Recently, a significant impact of changes in pupil size on gaze position as measured by camera-based eye trackers has been reported. In an attempt to improve the understanding of the magnitude and population-wise distribution of the pupil-size dependent shift in reported gaze position, we present the first collection of binocular pupil drift measurements recorded from 39 subjects. The pupil-size dependent shift varied greatly between subjects (from 0.3 to 5.2 deg of deviation, mean 2.6 deg), but also between the eyes of individual subjects (0.1 to 3.0 deg difference, mean difference 1.0 deg). We observed a wide range of drift direction, mostly downward and nasal. We demonstrate two methods to partially compensate the pupil-based shift using separate calibrations in pupil-constricted and pupil-dilated conditions, and evaluate an improved method of compensation based on individual look-up-tables, achieving up to 74% of compensation.

Show MeSH