Tagged: accuracy. light reflection
- 01/10/2014 at 02:54 #1769Stan YuParticipant
Hi all, I am trying to use the eyex controller to navigate the mouse in c#. However, one of the most critical issue for me is accuracy. While he device track the gaze point, it will behave some following issues.
1. It will not response for certain position on the screen (as my code is to move the mouse to the point that I am gazing, my monitor is 23′ size). e.g, it will not move further if I am trying to gaze on the area near the edge of screen.
2. It would jump to those coordinate that I was not gazing on. This would happen the same in the calibration screen (the one that has 9 yellow circles to verify the calibration effect), the tracking dots would occasionally jump to somewhere else that I was not looking at.
3. With or without glass on would have significantly different accuracy, without glass would behave way more accurate than with glass.
My guess is the following factors might cause the accuracy problem.
1. With glass on.
2. Reflection on the screen (If you are in a bright room without closing up the curtain, and light from window reflects on your screen)
Please advice my query and provide solution if there is any. I really want to explore the potential of this device.
Thank you.02/10/2014 at 09:15 #1785Mattias [Tobii]Participant
Hi Stan Yu,
You are spot on with your assumptions. Glasses do deteriorate accuracy and increases noise for a lot of users. Mind, not all users with glasses, but several. Same goes with direct sunlight or the precense of other equipment that emits light in the so called near infra red wavelength region. They can interfere.
So in short your experience that the accuracy improves when you take off your glasses follows the general trend. The stronger the glasses, the stronger the impact (typically). With that said a lot of people find that the EyeX Controller works remarkably well with glasses compared to other eye trackers.
And now on to some positive sidenotes. I appreciate your feedback. We are quite aware of this problem and are actively improving our algorithms and whenever there is a new firmwaer or EyeX engine up for download, you should check it out to see if you experience improvements. We are always happy for feedback.
Some brief comments on your observations 1 and 2.
1. Yes. Corners are typically most prone for accuracy offsets. However this is one area where we are right now focussing our development effort so look out for new releases.
2. Noise. With this sort of measuring system there will always be a risk of false detection, i.e noise, and in some cases the noise is big. You will however be happy to hear that this is another key focus area for our development team so stay tuned.
As the only real direct help I can give right now, you can check if your glasses are clean. Also if they are old and have a lot of scratches or marks it will deteriorate the quality of the eyetracking. But from your analysis I think that this is probably things you have already checked for.
Mattias02/10/2014 at 09:35 #1786Stan YuParticipant
Thank you very much , that is a really comprehensive reply.
I will keep checking in the update and forum to get the latest information.
Meanwhile, appreciate your hard work and I look forward to seeing the new improvement of this device in future.14/05/2015 at 00:02 #2972antoine amanieuxParticipant
exactly the same problem with eye tribe, the technology is simply not working. why don’t they keep it in the lab for a few more years until it is ready ?08/06/2015 at 12:36 #3077Jenny [Tobii]Participant
We are very aware that eye tracking today is not pixel-perfect, and that it probably never will be. But we have seen it proven more than once that a good user experience can be created based on the current accuracy and precision in combination with a user interface designed with eye-gaze interaction in mind. The point is that it takes something more than high accuracy and precision to create a good user experience with eye tracking. This is how one of my collegues have put this:
When creating interactive applications making use of eye tracking for the general public, a core concern for any developer should be the consistency of end user experiences. A natural, effortless and magical user experience that works for one person out of twenty can be an interesting concept at most. One that works for anyone will change the way your users interact with computers fundamentally.
Historically, one of the core metrics by which eye tracking devices have been measured is so called ideal accuracy and precision. This is defined as the best accuracy/precision measured for the most easy to track user. In many cases this is an interesting benchmark, especially within research where one could try several test participants before finding one that it works on. However ideal accuracy and precision gives no information of how often you will see this level of performance for a random user, making it much less useful as a metric for interactive applications using eye tracking.
For developers of interactive applications that are intended to be used by any user, focus should rather be on how consistent the end user experience will be over time, during movement, in varying surroundings and between different users.
- You must be logged in to reply to this topic.