I have been analysing smooth pursuit data for a few years now - I used
to collect it with a Skalar IRIS and a whole load of custom built
hardware/software to get the signal clean. THen I used a DOS program
called EYEMAP - made by Amtech GmBH - to actually analyse the data (e.g.
derive measures such as velocity gain / RMSE etc etc. The program was
written ages ago, and they no longer support it. It was quite buggy, and
the real problem was that it was not at all clear how the program was
deriving the measures it gave you - you had to trust it, and I never
really did... Now I have an eyelink eye tracker, and I'm beginning to
think about how to analyse pursuit data. It looks like I will probably have
to write something in MATLAB, because I don't think there are any
commercial programs designed for this. If I ever get anywhere, I'll
let you know - I'm a lecturer, and don't have much time unfortunately.
If you get any replies saying "I have the software to do this" please
forward them to me!


Your best bet is to look at some of the early smooth pursuit papers - I
don't have any refs to hand, but I've been glancing at a few recently.
These may have some reasonably detailed algorithms for blink detection
etc etc. One problem with analysing smooth pursuit is detecting the
intrusive saccades which occur during pursuit - saccade detection
normally works by checking for changes in velocity, but as velocity is
changing all the time during sinusoidal pursuit, this becomes very
tricky mathematically...

As I'm sure you'll find out - there are no "rules" for analysing smooth
pursuit data, and determining the "best" way of doing it has caused lots
of academic friction over the years. It all really depends on what your
questions are - e.g. if you want to know whether pursuit is impaired ni
a population or not, you could probably get away with a "qualitative"
judgment - simply rating the pursuit on a 0-5 scale on the basis of how
good it "looks". This is how people used to do it, and I still see it
occaisionally today. Obviously its not very scientific and now most
people try to use quantitative measures - often "global" measures such
as lnS/N and RMSE. Issues such as whether these measures shuold be taken
over the whole recording, or whether blinks / saccades should be removed
first etc etc, are all still debated. More meaningful measures are gain
(the ratio of eye velocity to target velocity), and for sinusoidal
pursuit, phase-lag. I also like to count and "classify" the saccadic
intrusions that occur during pursuit - these are usually catch-up
saccades, but you can also get square-wave jerks, backup saccades and
anticipatory saccades. Again, there are papers (Friedman I think) that
give guidelines on classification.

If you give me more details of your research, I may be able to point you
at some papers. One possible problem with the Eyelink set up is that it
only samples at 250Hz - this means that detection of saccadic intrusions
might be problematic - I'm not sure... Eyelink II can do 500Hz which
should be OK.

The more I play with the Eyelink software that displays teh pursuit
target, the less clear I am about exactly how accurate the timing
information is - it only sends a signal to the edf file with the targets
location every once in a while, and to analyse pursuit properly, you
really want to know exactly where the target was and exactly where the
eye was over as many samples as possible. It looks like I will have to
"extrapolate" the targets position between signals, and I'm not sure
this is ideal...


Hope this helps.

Sam.


Dr S.B.Hutton
Experimental Psychology
Biological Sciences
University of Sussex
Falmer
Brighton
UK


>
> Hello,
>
> I am a PhD student beginning to do work measuring accuracy of horizontal
> smooth pursuit to sinusoidally moving targets using an Eyelink tracker.
> I have no previous experience with this and am hoping to find some
> references giving detailed algorithms for getting fully from the point of
> having just the raw horizontal data to the point of calculating measures
> such as modal/peak gain and root mean square error (i.e., ideally
> incorporating strategies for dealing with general clean-up issues such as
> blink-removal).
>
> Can anyone recommend a reference or two to help me begin to get up to
> speed on this data-analytic aspect of things?
>
> Please just email me directly on this.
>
> Many thanks!
> Jeremy
>
> p.s. Also, I am trying to decide whether I need to use a bite-bar in order
> to get a viable measurement of horizontal smooth pursuit accuracy with the
> Eyelink.  Any advice on this?
>
>
> _________________________
> Jeremy Wilmer
> William James Hall - Room 710
> Vision Lab - Psychology Department
> Harvard University
> Cambridge, MA
> 617-493-7105
>
> --
> EYE-MOVEMENT mailing list ([log in to unmask])
> N.B. Replies are sent to the list, not the sender
> To unsubscribe, etc. see http://www.jiscmail.ac.uk/files/eye-movement/introduction.html
> Other queries to list owner at [log in to unmask]
>

--
EYE-MOVEMENT mailing list ([log in to unmask])
N.B. Replies are sent to the list, not the sender
To unsubscribe, etc. see http://www.jiscmail.ac.uk/files/eye-movement/introduction.html
Other queries to list owner at [log in to unmask]