The Eye Tribe

Hi,

I want to share with you, something that I realized. I realized that the background color of the screen is an important variable about the eye gaze validity. I used at the same application as background colors, the white or black respectively. The results is totally different. I think, the reflection of the room light on the screen has a high impact on the proper operation of the device.

Giannis

Hi Giannis,

may you share with one did you get better results?

With the black as background color.

.

Correct.

Suggested protocol is to calibrate to the same color, or rather luminance, as the stimuli/UI in terms of pupil response.

Edit: Black is best, larger pupil gives more accurate measurement.

I wrote a little note with these results:

Significant variances between participants, calibration backgrounds, and test backgrounds were observed. The least mean systematic error (deviation of recorded gaze position from target) was obtained when the calibration background and test background were black (27 pixels). Systematic error increased when displaying a test background that deviated from the calibration background intensity. Hence, the largest mean systematic error occurred when calibrating to a black background and displaying a white background (67 pixels). For complex chest CT volumes the white calibration background performed best (38 pixels). An angular analysis of the systematic error was performed and demonstrated that the systemic error primarily affects the vertical position of the estimated gaze position.

Accuracy of a remote eye tracker for radiologic observer studies: effects of calibration and recording environment

I think that I understood better my mistake.

To be more specific:

for black color the r.g.b. is (0, 0, 0)
for white color the r.g.b. is (200, 200, 200) // light gray

During the calibration phase I used the black color as background, but during my application I used the light gray color. The result was poor, because the gaze was shifted up on Y axis.

Under these conditions, I did the below test. I created 4 conditions, with 3 trials per condition.

A)
Calibration Phase : light gray --------------> Application Phase : light gray
Results:
* Difficulty to get Excellent calibration.
* Very Good gaze.

B)
Calibration Phase : light gray --------------> Application Phase : black
Results:
* Difficulty to get Excellent calibration.
* Poor gaze. High downward shift on the Y axis.

C)
Calibration Phase : black --------------> Application Phase : light gray
Results:
* Easy to get Excellent calibration.
* Not Good gaze. Middle upward shift on the Y axis.

D)
Calibration Phase : black --------------> Application Phase : black
Results:
* Easy to get Excellent calibration.
* Very Good gaze. (Maybe at first condition the result was little better than fourth condition.)

Thanks Martin!!

Also, I adjusted a little the brightness and the contrast of my screen.

The final result is so GOOD...

I couldn't understand why I had poorly results.

Now, I know...

I feel a relief

Martin wrote:Correct.

Suggested protocol is to calibrate to the same color, or rather luminance, as the stimuli/UI in terms of pupil response.

Edit: Black is best, larger pupil gives more accurate measurement.

Dear Martin, dear all,

larger pupil equals more pixels available to determine pupil center, hence, more accurate estimation of the pupil center. In practical application, I nevertheless prefer to reduce pupil size with a bright screen and some ambient light.

My considerations:
- the noise level should theoretically be dominated by the much smaller corneal reflex anyway.
- a larger pupil is more likely to be occluded by droopy eyelids, resulting in nonsensical pupil center estimations.
- the chance of droopy eyelids because the subject gets tired is also lower in a bright environment.

I would be interested in your opinion, especially with respect to the first point, the high noise due to the small CR and how that relates to the quality of the pupil center estimate.

Best regards