WEBVTT

00:00:00.000 --> 00:00:00.680 align:middle line:90%


00:00:00.680 --> 00:00:02.810 align:middle line:84%
So, Laurie, you've got
these nifty glasses on.

00:00:02.810 --> 00:00:06.140 align:middle line:84%
I guess that means it's
time for preprocessing.

00:00:06.140 --> 00:00:07.490 align:middle line:90%
Yes, it does

00:00:07.490 --> 00:00:10.220 align:middle line:84%
Yeah, so can you tell me
a little bit about what

00:00:10.220 --> 00:00:13.080 align:middle line:90%
preprocessing is, why we do it?

00:00:13.080 --> 00:00:15.110 align:middle line:84%
What are the big
things to look out for?

00:00:15.110 --> 00:00:18.050 align:middle line:84%
Yeah, so preprocessing
is important

00:00:18.050 --> 00:00:21.950 align:middle line:84%
because this is the step in
which you get rid of artefacts

00:00:21.950 --> 00:00:22.940 align:middle line:90%
from your data.

00:00:22.940 --> 00:00:26.480 align:middle line:84%
Usually these are
attributable to either blinks

00:00:26.480 --> 00:00:29.690 align:middle line:84%
or to instances
of poor tracking.

00:00:29.690 --> 00:00:33.860 align:middle line:84%
For example, if the participant
looks at a very extreme angle.

00:00:33.860 --> 00:00:37.650 align:middle line:84%
Yeah, so you're hooked
up to the computer here.

00:00:37.650 --> 00:00:40.940 align:middle line:84%
Can you show me what lost
data would look like?

00:00:40.940 --> 00:00:43.250 align:middle line:84%
For example, if you're
not looking at the thing

00:00:43.250 --> 00:00:44.720 align:middle line:84%
you're supposed
to be looking at.

00:00:44.720 --> 00:00:48.560 align:middle line:84%
Yes, so if I'm looking at a
very extreme angle right now--

00:00:48.560 --> 00:00:49.090 align:middle line:90%
Ah, yes--

00:00:49.090 --> 00:00:51.080 align:middle line:84%
--have you lost track
of where my pupils are?

00:00:51.080 --> 00:00:52.190 align:middle line:90%
Yeah.

00:00:52.190 --> 00:00:54.830 align:middle line:84%
It's getting a lot of yellow
in that one eye there.

00:00:54.830 --> 00:00:56.400 align:middle line:90%
That's probably not good.

00:00:56.400 --> 00:00:58.220 align:middle line:84%
So we have to remove
that sort of data.

00:00:58.220 --> 00:00:58.760 align:middle line:90%
Yeah.

00:00:58.760 --> 00:01:02.700 align:middle line:90%
And also if I were to blink.

00:01:02.700 --> 00:01:03.570 align:middle line:90%
Yeah.

00:01:03.570 --> 00:01:05.580 align:middle line:90%
You can see it disappears.

00:01:05.580 --> 00:01:08.160 align:middle line:84%
And I don't know if you can
see this very well right now

00:01:08.160 --> 00:01:12.780 align:middle line:84%
but it actually takes a fraction
of a second for the tracking

00:01:12.780 --> 00:01:16.930 align:middle line:84%
to re-establish where the
pupil is after a blink.

00:01:16.930 --> 00:01:17.430 align:middle line:90%
Yeah.

00:01:17.430 --> 00:01:20.910 align:middle line:84%
And then another thing is
if you can slowly blink,

00:01:20.910 --> 00:01:25.470 align:middle line:84%
if that makes sense, maybe we
can see examples of these edge

00:01:25.470 --> 00:01:26.530 align:middle line:90%
artefacts.

00:01:26.530 --> 00:01:31.420 align:middle line:84%
So the pupil tracker has a
very, very high sampling rate.

00:01:31.420 --> 00:01:35.430 align:middle line:84%
And so you can see it'll
register pupil sizes that

00:01:35.430 --> 00:01:39.090 align:middle line:84%
are unrealistically small before
the blink has fully completed

00:01:39.090 --> 00:01:40.710 align:middle line:90%
and the eye is fully closed.

00:01:40.710 --> 00:01:43.440 align:middle line:84%
So maybe, yeah
and then, see, you

00:01:43.440 --> 00:01:46.950 align:middle line:84%
can see that the pupil
is still being registered

00:01:46.950 --> 00:01:48.540 align:middle line:90%
but with a very small size.

00:01:48.540 --> 00:01:49.980 align:middle line:90%
It's mostly occluded.

00:01:49.980 --> 00:01:52.900 align:middle line:84%
And so we want to get
rid of that, as well.

00:01:52.900 --> 00:01:58.410 align:middle line:84%
Personally, I use preprocessing
scripts in R. What about you?

00:01:58.410 --> 00:02:00.090 align:middle line:84%
Usually that's what
I use, as well.

00:02:00.090 --> 00:02:01.380 align:middle line:90%
There are alternatives.

00:02:01.380 --> 00:02:04.440 align:middle line:84%
Most of the
eye-tracking software

00:02:04.440 --> 00:02:07.290 align:middle line:84%
also has algorithms
within the software

00:02:07.290 --> 00:02:09.699 align:middle line:84%
that you can use to
do this automatically.

00:02:09.699 --> 00:02:13.110 align:middle line:84%
But for me, I prefer to have
control over every aspect

00:02:13.110 --> 00:02:14.160 align:middle line:90%
of this process.

00:02:14.160 --> 00:02:18.510 align:middle line:84%
Yeah, I agree 100% And so what
are some of the data checks

00:02:18.510 --> 00:02:20.160 align:middle line:90%
that you go through?

00:02:20.160 --> 00:02:23.280 align:middle line:84%
Assuming you've scrubbed
most of the blanks out now,

00:02:23.280 --> 00:02:25.140 align:middle line:90%
you're left with cleaner data.

00:02:25.140 --> 00:02:26.850 align:middle line:84%
But you just want to
make sure that it's

00:02:26.850 --> 00:02:28.890 align:middle line:84%
ready to go for
the full analysis.

00:02:28.890 --> 00:02:32.370 align:middle line:84%
Well, for me, usually, I would
look at the curve, actually

00:02:32.370 --> 00:02:34.980 align:middle line:84%
the time series data
of the pupil curve,

00:02:34.980 --> 00:02:38.460 align:middle line:84%
and make sure that there
are no very extreme values.

00:02:38.460 --> 00:02:45.420 align:middle line:84%
And I try to see if there are
very fast jumps in the signal

00:02:45.420 --> 00:02:50.390 align:middle line:84%
because, of course, the eye
does not move all that fast.

00:02:50.390 --> 00:02:53.190 align:middle line:84%
And I look for,
also, gaps caused

00:02:53.190 --> 00:02:56.460 align:middle line:84%
by long periods of either
having the eyes shut

00:02:56.460 --> 00:02:58.440 align:middle line:84%
or long periods
of poor tracking.

00:02:58.440 --> 00:03:00.900 align:middle line:84%
Yeah, and with
some of these gaps

00:03:00.900 --> 00:03:04.980 align:middle line:84%
you can take a pretty
good stab at what

00:03:04.980 --> 00:03:08.620 align:middle line:84%
the pupil would look like if
you had tracked that window.

00:03:08.620 --> 00:03:09.180 align:middle line:90%
Yeah.

00:03:09.180 --> 00:03:10.722 align:middle line:84%
At least when they're
smaller, right?

00:03:10.722 --> 00:03:14.190 align:middle line:84%
Yeah, there are a number
of gap-filling algorithms

00:03:14.190 --> 00:03:15.167 align:middle line:90%
that you can use.

00:03:15.167 --> 00:03:17.250 align:middle line:84%
But, of course, you want
to be careful with these.

00:03:17.250 --> 00:03:19.480 align:middle line:84%
You don't want to fill
gaps that are too long.

00:03:19.480 --> 00:03:19.980 align:middle line:90%
Yeah.

00:03:19.980 --> 00:03:22.230 align:middle line:84%
And, also, it depends
on what you're

00:03:22.230 --> 00:03:23.460 align:middle line:90%
planning to do with the data.

00:03:23.460 --> 00:03:28.470 align:middle line:90%


00:03:28.470 --> 00:03:31.180 align:middle line:84%
If it's going to do a very fine
grained analysis of the data,

00:03:31.180 --> 00:03:35.610 align:middle line:84%
you don't want to fill with not
real data, very small amounts

00:03:35.610 --> 00:03:37.620 align:middle line:84%
of time that you're
going to look up.

00:03:37.620 --> 00:03:40.740 align:middle line:84%
Yeah, the more
missing data there

00:03:40.740 --> 00:03:44.360 align:middle line:84%
is, the harder it is to make
an educated guess about what

00:03:44.360 --> 00:03:46.110 align:middle line:84%
it actually would have
looked like had you

00:03:46.110 --> 00:03:47.027 align:middle line:90%
been able to track it.

00:03:47.027 --> 00:03:48.000 align:middle line:90%
Yes.

00:03:48.000 --> 00:03:48.880 align:middle line:90%
So that's cool. .

00:03:48.880 --> 00:03:49.380 align:middle line:90%
Yeah.

00:03:49.380 --> 00:03:54.300 align:middle line:84%
So in summary, the main goal
here is to remove blinks.

00:03:54.300 --> 00:03:56.040 align:middle line:90%
To remove blinks and--

00:03:56.040 --> 00:03:57.640 align:middle line:90%
Check errors.

00:03:57.640 --> 00:03:58.410 align:middle line:90%
Yes

00:03:58.410 --> 00:04:01.220 align:middle line:90%
OK, all right.

00:04:01.220 --> 00:04:12.000 align:middle line:90%