Home › Forums › Feature Requests › Math fine-tuning
- This topic has 3 replies, 3 voices, and was last updated 9 years, 1 month ago by gigertron.
- 23/01/2014 at 07:52 #287gigertronParticipant
It’d be nice to be able to fine-tune the filtering for the gaze data and fixation point data. I’ve been able to get very usable results from the Rex by throwing a damped response filter on the gaze data and giving the output of that a dead zone, but it’d be nice to be able to just set those values as parameters in the API.
I’d also like to know what algorithms are used to determine fixation points – and be able to fine tune and bias the noise and threshold values for fixation.
The sample projects are very clean and easy to work with – no complaints about these : D23/01/2014 at 12:16 #290Jenny [Tobii]Participant
Thanks for your feedback!
It’d be nice to be able to fine-tune the filtering for the gaze data and fixation point data.
We’ll take this into consideration as the API evolves.
I’d also like to know what algorithms are used to determine fixation points
Yes, hopefully this will be clearer when the documention is in place. We are working on it 🙂
The sample projects are very clean and easy to work with – no complaints about these : D
I’m very glad to hear that! 🙂 We’ve worked hard to make them simple yet useful.14/02/2014 at 07:30 #463Cheng GuoParticipant
would you mind to explain the “damped response filter” you have implemented?
I would like to reduce the jittering of the eye tracker data. I tried to store up to 10 most recent points and average them. It kinda works but I would like to explore other possibilities.
Thanks!14/02/2014 at 08:59 #464gigertronParticipant
If you/ve done signal processing theory it’s obvious, otherwise get ready for some math!
Put simply: we start with some location, the current one.
A new point comes in, and we find the ‘error’ – meaning, the x/y distance from where we are to where the new point is.
Then, we move our point towards the input by some percentage of the error. This can be anywhere from 0 to 1, with values like 0.1 or 0.8 being the most useful.
Then you just apply this, one point at a time. really simple, but it’ll wobble as much as the weighted average will. So I apply a deadzone after that. Play around with it and see what works for you. I’ll probably post my code this weekend.
- You must be logged in to reply to this topic.