Hi,

I have not used the pupillometry data from my eye tracking recordings but I can assume that you will want to ignore the data when the participants were blinking. I recommend R's saccade package for eyeblink/saccade/fixation detection: https://github.com/tmalsburg/saccades That's what I used and the performance was better than with Tobii's software.

To use rsaccade, I had to convert the json data (procuded in my case by Tobii Pro Glasses 2) to a regular tsv/csv data file. For that I used the preprocessing python script in Jeff MacInnes' Mobile Gaze Mapping: https://github.com/jeffmacinnes/mobileGazeMapping/tree/master/preprocessing You could probably adapt that or code one in R.

A mild annoyance with the saccade package's format.gazedata() function is that it wants the data in a very specific format that uses only one column to identify the participant & trial. There can be no NAs in timestamps or coordinates, and the time stamp has to be in chronological order or the function won't work. Here's what I did. Here timeStamp is exactly that, ref_gazeX and ref_gazeY are the coordinates as recorded by the Tobii device, participant is participant id and question is the item id.

-----

library("data.table")

library(tidyverse)

#library("devtools")

#install_github("tmalsburg/saccades/saccades", dependencies=TRUE)

library(saccades)

format.gazedata

This opensource is in Matlab but you may find it useful:

https://link.springer.com/article/10.3758/s13428-018-01190-1

You can also try this R open source software for pupillometry analysis:

https://github.com/samhforbes/PupillometryR

Thanks for all these great tips!

The pupil signal may contain some noise which you can remove using moving average filter. This can also help in filling missing values.

You can take a look at the pupil signal processing in this paper: Article Pupillometry Based Real-Time Monitoring of Operator’s Cognit...