Drift Direction DistributionN=22

0

1

2

3

-1

-2

-3

0

1

2

3

-1

-2

-3

0-1-2-3 1 2 3 0-1-2-3 1 2 3Horizontal drift (deg) Horizontal drift (deg)

Hor

izon

tal d

rift

(deg

)

Introduction Methods

Conclusions

A binocular evaluation of pupil-size dependent deviation in measured gaze position

1) Center for Mind/Brain Sciences, University of Trento, Rovereto, Italy 2) Centre for Vision Research, York University, Toronto, ON, Canada 3) Kunming Institute of Zoology, Yunnan University, Kunming, China

J. Drewes1,2, W. Zhu3, Y. Li3, Y. Hu3, F. Yang3, X. Du3, X. Hu3

contact: mail@jandrewes.de

Results

2s 2s 2s 2s 2s

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 [Wyatt 2010], and a first attempt at compensating for this drift was proposed [Drewes et. al. 2012].

Re

port

ed

ve

rtic

al g

aze

po

sitio

n (

de

g)

Rep

ort

ed

ho

rizo

nta

l gaz

e p

osi

tion

(de

g)

Time from trial start (ms) Time from trial start (ms)

A1 A2 B1 B2

1500 2000 2500 3000 3500 4000 1500 2000 2500 3000 3500 4000 1500 2000 2500 3000 3500 4000 1500 2000 2500 3000 3500 4000

2

1.5

1

0.5

0

-0.5

-1

-1.5

-2

2

1.5

1

0.5

0

-0.5

-1

-1.5

-2

2

1.5

1

0.5

0

-0.5

-1

-1.5

-2

2

1.5

1

0.5

0

-0.5

-1

-1.5

-2

Time from SO (ms)

Mea

sure

d ga

ze p

ositi

on (

deg

)

Paradigm

To improve 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 20 subjects. Subjects were students at Yunnan University. Eye movement data was recorded binocularly using an SR Research Eyelink 1000. Pupil size changes were induced by changing the background luminance of the screen. Gaze measurements were calibrated on 5x5 point grids with 5 degree spacing, using several background luminance levels. We evaluate 3 different approaches to compensate for the pupil-size induced drift.

The absolute pupil-size dependent drift varied greatly between subjects (0.8 to 4.3 degree, mean 2.4 degree), but also between the eyes of individual subjects (0.16 to 1.7 degree difference)

The preferred drift direction was inwards and down, in direction of pupil constriction

Our compensation method optimized in direction of constriction reduced drift by 72% (avg).

Calibration at constant luminance levels Drift measurement at variable luminance

While ground truth was presented in Drewes et al., all previous studies used very few subjects to demonstrate this effect (5 and 2 respectively), and only monocular measurements were performed. Drewes, J., Masson, G. S., & Montagnini, A. (2012). Shifts in reported gaze position due to changes in pupil size: ground truth and compensation. In Proceedings of the Symposium on Eye Tracking Research and Applications (pp. 209–212). New York, NY, USA: ACM. doi:10.1145/2168556.2168596

Wyatt, H. J. (2010). The human pupil and the use of video-based eyetrackers. Vision Research, 50(19), 1982–1988. doi:10.1016/j.visres.2010.07.008

Raw Data

N=26

0 2 4 6 8 10 0 2 4 6 8 10Time (s) Time (s)

Average Residual Error:

2-Point: 43%3-Point: 40%LUT: 28%

Residual Error per Time-Interval

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

0

-1

-2

-3

-4

1

2

3

4

Mea

sure

d g

aze

pos

itio

n (d

eg)

Mea

sure

d g

aze

pos

itio

n (d

eg)

Mea

sure

d g

aze

pos

itio

n (d

eg)

Subjects were asked to fixate steadily, while background luminance changed

Horizontal (X) Vertical (Y)S1 left eye S1 left eye

S1 right eye S1 right eye

S2 right eye S2 right eye

Drift Compensation2-Point: Gaze data is interpolated between 2 calibrations at maximum and minimum luminance, with pupil size as interpolation index. Fast, simple, but only linear – not very precise.

3-Point: Gaze data is interpolated by fitting a quadratic function to 3 calibrations at maximum, minimum and intermediate luminance, with pupil size as interpolation index. Still fast, more accurate than 2-Point.

LUT: Gaze data is corrected by means of a look-up-table mapping pupil size to drift magnitude. The LUT is constructed in direction of pupil constriction, resulting in best accuracy during constriction, but reduced accuracy during dilation (hysteresis).

Raw gaze

2-Point

3-Point

LUT

S1 left eye S1 left eye

S1 right eye S1 right eye

S3 left eye S3 left eye

S4 right eye S4 right eye

0 2 4 6 8 100 2 4 6 8 100 2 4 6 8 100 2 4 6 8 10

-1

0 2 4 6 8 10

1

2

0

-1

1

2

0

-2

2

4

0

-2

2

4

0

-2

2

4

0

-2

2

4

0

-1

1

2

0

-1

1

2

0

3 3

Me

asur

ed g

aze

(deg

)M

eas

ured

gaz

e (d

eg)

Horizontal (X) Vertical (Y) Horizontal (X) Vertical (Y)

Drift Compensation Results

0-2s 2-4s 4-6s 6-8s 8-10s

Time (s) Time (s) Time (s) Time (s)

Mean drift amplitude: 2.4 degree (0.8 – 4.3)

0

0.5

1.0

1.5

Ave

rag

e d

rift

ma

gnitu

de

(deg

)