I am working with eye movement metrics to infer visual task based on observed eye movement patterns.
One of these metrics is total scanpath length. Based on what I have read, I need to find the length of movement (saccade) between consecutive fixations. I assumed that this can be easily calculated using Euclidean distance in pixels, then converting it into degree of visual angle. Is my assumption correct?
When I searched how a scanpath is computed, I encountered some programs, where the scanpath is compressed and also similarity between scanpaths is calculated. Do I need to compress a scanpath (by removing repetitive sequence) in order to correctly calculate its length? For what purpose the similarity between two scanpaths is used?
Are they any additional criteria that I should also consider?
I am not 100%ly sure for which you need the measure, but to calculate the length of a saccade (and so the total length) is pretty simple. Just take the square root of the sum of squares of x and y distances for each saccade and then sum them up.
About the similarity: which kind of routine do you have in mind here, because this is pretty complex, in contrary, especially as the number of fixations does differ across subjects of course.
Best wishes,
CCC
Thank you for your prompt answer.
I don't have a specific use in mind, but I read similarity of scanpaths can be used to show the difference between users perfoming a specfic visual task such visual search. Is it possible to use it to differ between two completely different visual tasks?
I suppose that it is better to calulate the distance between main scan-path only. I mean there are specific points of an image which attract maximal attention of a subject ("attractors")
Hey Jihad!
There are is a myriad of different ways of comparing scanpaths which result in different measures and require different preprocessing stages. A good overview on measures you can use and there specific dis-/advantages is given in Holmqvist's book (chapter III section 10/11).
If you have a fixed amount of viewing time (and negligible data loss) for each participant, euclidean distance will serve you well. But this also depends on your research question (e.g. are you actually interested in the travel distance or is it more important to measure the area covered by the scan path).
For comparing scanpaths between different tasks (or views) I would recommend some more elaborate metric (e.g. recurrence quantification, multi-match).
Hope that helps, Greetings, David
Article Eye Tracking: A Comprehensive Guide To Methods And Measures
Thank you David for the book recommendation.
I am actually interested in calculating both, i.e. total traveled distance and area covered by the scan path. The viewing time is currently fixed, so I think the Euclidean distance is still a valid way?
Hey Jihad,
if viewing time is fixed, you are indeed fine....
but
You should take care of missing samples (e.g. due to blinks and other artifacts), as you can not compute the distance from them but they still contribute to the number of samples. Consider interpolating these values (e.g. with a spline) which will be give reasonable data.
Greetings, David
Hello David,
I have a question, using an eye tracker, how do you know some samples were missing during recording?
Cause when I calibrate, I can know how many samples were used from the total samples to know if a calibration is good or not from eye tracker API. I also get fixation-related info that I can use for later calculations.
Thanks
If you are still interested in scanpath comparison, you can check our tool www.eyetracking.upol.cz/scangraph
- Badges
- Science topic
- Similar topics
- Psychology