Hacker News Reader

Comment by natebleker | Hacker News Reader

Disclaimer, I am the primary engineer of a commercial eye tracking system.

Tools like these popup every now and then are a nice tool for rough estimation of gaze. Fixation tracking is a large complicated issue that usually requires some sort of calibration to get more precise results. By sidestepping the calibration problem, much higher subject compliance can be achieved since you don't have a grad student barking confusing order at you. The downside to this is the noise you see in the tracking results. For those interested a product that produces similar results is pupil labs ambient gaze tracking "Core" research headset[0]. [0] https://pupil-labs.com/products/core/

View on HN

Author here (I didn't use Show HN because it's a larger collaborative project). Yes this is basically a tradeoff between accuracy and flexibility+accessibility.

For usability testing, physical eye trackers have to be used one person at a time (no simultaneous use), use an experimenter's time to schedule and administer, can only be used for a short period, and only work with local participants. But yes they will probably always have better accuracy, and detect saccades/fixations better which are also great for psychology studies.

The other thing is if you want to make a consumer application (like browser game, or accessibility mode), then it's more practical to have people just consent to having their webcam turned on, than to go out and buy an eye tracker.

In your estimation, how far away are we from eye-tracking software being able to detect the start of a microsaccade, estimate where the gaze is moving toward, and draw new contents on the screen there before the gaze even reaches that point? I would think that by "hacking" into human brain "vulnerabilities" like saccadic masking and chronostasis, such software could potentially yield seriously trippy and mind-altering results!

"How far away are we" is a strange question to ask about something we already do. You can read about this technique being used to investigate how we read here: https://docs.microsoft.com/en-us/typography/develop/word-rec...

Note the citation to a study that used the technique in 1975.

There’s two parts to that implementation of such a system, and they’re both interesting! The first part is detection of a micro saccade which is already available in research systems I have personally worked on. You can basically crank up the camera frame rate until you’re around the micro saccade range and do some clustering analysis on the positional data to decouple the hardware moving vs the face. The second part is having a fast enough reliable commercial grade displa system to present stimulus on. Displays are very fickle in practice and getting your hands on one that can run with adequate color, contrast, and brightness at speed is currently very difficult. One angle under research currently is the perception of stimulus during saccades and micro saccades. There’s quite a bit of time and effort in the industry going into neurological assessments through saccades, the tools coming out of this are really starting to come down in price. This opens the door for a bunch of lower priced research options, such as the parent article, to enable a much more rapid pace of understanding.

That's fascinating, but I'm wondering what is the practical value of detecting microsaccades? Aren't they just involuntary twitching?

They’re looking to be increasingly more interesting as a biomarker for MS among other things [0]. There’s a lot we still don’t know about our visual systems and the advent of lower cost hardware is allowing huge amount of interesting questions to be hypothesized and tested. It’s a great time to be in the science behind all of this, like the authors at Brown on this paper!

[0] https://iovs.arvojournals.org/article.aspx?articleid=2670267

For the duration of the microsaccade, you're blind. So if something changes onscreen it's much harder to see.

IIRC people have used small orientation changes during microsaccades for redirected walking in VR. You feel like you're walking straight but you're actually curving back on yourself.

Edit: I think that was just detecting full-on saccades but a microsaccade version would be smoother and harder to detect.

Source: https://blog.siggraph.org/2018/05/challenge-accepted-infinit...

I believe this might be available in less than 5 years, potentially less than 2, provided both http://www.adhawkmicrosystems.com/ and https://www.kura.tech/ ship and partner with each other. If either one of those companies dies or their technology doesn't work out like I hope, then it might be a while longer.

It depends on your application. If you're trying to replace your mouse with gaze, then yes you need careful calibration. I have worked on some purpose-built gaze interactive experiences and coarse accuracy without calibration can be enough to do something useful.

What are your thoughts on newer ML based approaches like MITs "Eye Tracking for Everyone"?

https://gazecapture.csail.mit.edu/

Reply on news.ycombinator.com