GazeHawk (YC S10) Does Eyetracking With Web Cams | Hacker News Reader

GazeHawk (YC S10) Does Eyetracking With Web Cams

techcrunch.com

149 pointsbwaldorf41 comments

About 2 years ago an acquaintance of mine suddenly had a stroke. When she came to she was conscious but completely paralyzed. The only way could communicate was with her eyes [1].

She's since recovered and is able to talk. But she describes an intense loneliness and frustration at being a prisoner in her own body. When she was finally able to use an expensive, specialized computer to communicate by slightly moving her fingers, it was an amazing encouragement.

Part of the problem right now is that these input systems are extremely expensive and specialized. Even if the family can afford to pay for one, they still have to spend time researching which system is the best.

I imagine a free, open-source system that will allow for rapid eyetracking-based writing. Say, a live cd that contains many commercial webcam drivers and boots into a system like Dasher (http://www.inference.phy.cam.ac.uk/dasher/DasherSummary2.htm...).

This would provide instant communication, however crude, to a person who has recently been through a tragic experience.

Is webcam eye-tracking imperfect? Yes. Do commercial systems exist that solve this problem better? Yes. But the idea here is to have something someone could put into hardware they already have to get _something_ working right away.

OpenGazer: http://www.inference.phy.cam.ac.uk/opengazer/ A nice list of eye trackers: http://www.cogain.org/wiki/Eye_Trackers#Open_source_gaze_tra...

*Yes, someone would have to help the paralyzed put the cd into the tray and boot the computer

[1]: http://www.caringbridge.org/visit/conniemchaddad/mystory

Props for linking to OpenGazer and Dasher. AEGIS is absolutely a terrific project, and I've been inspired by the work that Emli-Mari Nel and others are doing on exactly this problem.

Something to keep in mind when you're thinking about eye tracking for accessibility is how much custom hardware can greatly enhance the experience. Even if you're just mounting a webcam on a head bracket and removing the IR filter, some inexpensive work can really improve the user interface. (Of course, mounting a camera on your face isn't necessarily conducive to unbiased usability testing!)

One of my greatest hopes, personally, is that our approach to webcam-based eye tracking will yield algorithms and methods that we can give back to the community. Eye tracking has its shortcomings, to be sure, but the potential benefit is too great to be ignored.

You should check out the EyeWriter Initiaiative. I found the videos really inspirational.

http://www.eyewriter.org/

Is there an example page where we can test the accuracy of the eye tracking software?

I like the idea, but for $490, I'd like some assurance that the eye tracking technology actually works and that it's not just tracking the mouse or something else.

Haha, a reasonable request. We're working on a good way to let people demo our technology. Part of the problem is that this is computationally intensive, and we didn't want all of TC killing our processing servers.

The other part is we found that people demoing the software are less likely to follow the instructions (since they're not getting paid) and we worried about having a lot of poor tracks that would look bad. Our testers have been amazingly good at following instructions :).

It's only computationally intensive because you're using the processor on your end to do the work. Why not have a large portion of the work done on the clients machine?

Two main reasons:

1) Our processing can't be done realtime (i.e. it takes more than 1 second per second of video), at least for now, so there would need to be a lag that can get significant if they user has a slow enough computer.

2) We currently don't require any software to be installed client-side (we use Flash + JS), which we consider a big perk.

I can see the advantage of not having to install client side stuff, but if that's such a performance penalty that you have to charge so much more you're opening yourself up to competition by someone that does the same thing but finds a way to do it client side.

Looks like you have a bunch of tough optimisation ahead, much good luck with that.

Very nice idea by the way, eye tracking used to be the domain of specialised setups using cameras behind half-transparent mirrors, it's quite impressive that you've gotten this far with just a cheap webcam as an input device.

Maybe you could offer to pay your testers more if they're willing to do the processing (which requires a download). Depending on how much your processing costs are you could even distribute it @home style.

Have you tried to employ Graphics Processing Units (GPUs) to do the computation-intensive tasks?

I could imagine that using CUDA-enabled kernels might be of help for processing images.

Having done a bunch of user studies in the past, one of the big issues I've seen is the Hawthorne Effect ( http://en.wikipedia.org/wiki/Hawthorne_effect ) -- just the fact that the user knows they are being watched, they change their behavior so much that the study is irrelevant.

I think GazeHawk has the chance to push usability closer to being unbiased -- very promising!

Thanks! We didn't even realize this was one of the extra benefits of our technology, but it seems to be a pretty important one: people have reported completely forgetting that they're being tracked when using our early prototypes.

You can now do something very similar on the iPhone 4 because of the front facing camera. Imagine if you can hit-track when someone is looking at an ad. You could have an advertising business model at cost-per-glance. This works great if the ads are mainly for branding impact.

Yes. Very yes. I can't upvote this comment enough.

Advertising pricing with cost-per-glance (or eyeball, or something) is a big dream of ours. Front-facing cameras on mobile devices excite us no end.

Would it work with such a small screen?

People typically hold the iPhone closer to their face than they hold their laptop. Since accuracy in eye tracking depends on the angle between the calculated gaze vector and the true gaze vector, this means that error will be reduced.

The amazingly high pixel density on the iPhone 4 means that that the calculated gaze focus may be more pixels away from the real point - but we're working on some ways to compensate for that.

I'm surprised how most commenters here are willing to accept the idea of Gazehawk without questioning it. No disrespect meant to Brian and Joe, but any experienced researcher knows that any new research technology / method needs to be welcomed with a healthy cynicism until its value has been proven in the real world.

Firstly, can it really be as accurate as commercial eye tracking hardware? We're talking about the difference between $40,000 and $40 here - a Pepsi challenge is needed! Some leading figures in the UX industry claim that even the top end devices can easily become de-calibrated and provide bogus data.

Secondly, even if the technology is acceptably accurate, is the gazehawk research method effective at delivering findings that actually help you improve the design of your sites?

The idea of a predefined panel of 'users' is rather worrying. A user is defined as someone who actually uses your site. In the standard gazehawk offering, they provide a group of trained testers (who get paid $8 a pop), who may simply not care about the problem your site tries to solve. For example, if you have a webapp about mountain biking, you're going to get a panel of testers who may simply not care nor understand about types of bike, trails, nor understand any of the terminology you use. Eye tracking heatmaps are a product of conscious thought (as well as low level visual processing) - if testers don't care about nor understand the problem your site is trying to solve, the heatmaps you'll get from them will be almost entirely worthless.

Thirdly, analysis of eye-tracking data is very tricky. There are a lot of mistakes an untrained analyst will make, which could have major repercussions. I did a presentation on this at User Experience Lisbon a few months ago:

http://www.slideshare.net/harrybr/what-you-need-to-know-abou...

I really don't mean Brian and Joe any ill-will. All of these concerns can be dealt with if they share case study projects and having an open channel of discussion with the UX research community.

It's a shame this is for webpages only. I would very much like "focus follows eyes" in my window manager.

Yes, if it could actually be made to work accurately (down to the few-pixel level) and in real time then it would be a brilliant user interface element. Why should I click a mouse when I can just stare at the "reply" button for half a second?

(Yes, all sorts of potential badness which could happen if implemented unwisely.)

There is a big problem with eye-controlled interfaces called "Midas touch".. It gets extremely tiring to force yourself to look/not look in one point.

Looks like you beat us to the punch with the article!

Questions and comments are welcome. Thanks everyone!

Any more info about the actual test users? Are they using their own computers, or do they come to a test lab?

They use their own computers and webcams, providing a much more natural environment for testing. That's one of the reasons we can offer such a low-cost service.

If anyone is interested in being one of our users, sign up at https://www.gazehawk.com/tester/signup/

Sounds like an amazing idea... strikingly similar to the idea I pitched to YC in an interview in November, 2006 :-/. I guess they didn't like our team or progress. Bummer.

Brilliant.

we spend all day thinking about conversion rate optimization via conversionvoodoo.com - some of our larger clients have full hardware rigs and I can tell you this data is invaluable WHEN it's accurate.

Certainly moreso than usertesting due to the speed at which the eye moves / tracks - killer product, great release and can't wait to test it - check your site support email, I'd like to see how we could leverage this for more of our clients & case studies.

Very cool software and I would LOVE LOVE LOVE to use this, HOWEVER we make a hardware appliance and we can't let people on it (via web access) to do testing. I'd rather pay like $5k for a webcam and a software package to install and run tests with my customers (which are NOT normal web users, they are security experts) to see how they view our interface.

Thanks for the feedback. Right now we're targeting web-based interfaces, but expanding into different products is definitely something we're considering. It depends on what kind of demand we see for it.

The product portion I want to use is a web-based interface, however it can only be used for internal network use and we wouldn't be allowed to make it internet-accessible.

I think we should be able to webcam/monitor our users here, have an app gather data on our computers, then have our computers (which can access the internet AND our internal network appliance's interface which we are testing) upload the data to your website to be parsed.

I see. We may be able to work something out, email me at brian AT gazehawk.com and we can discuss the details. Our upcoming plans to let you use your own users may solve this.

Very impressive. They deserve to succeed for being so clever.

But are there more samples/examples? I can't find any on the site.

At $5 per site viewed, I hope the qualify the viewers! (oh wait it's "up to" $5)

( https://www.gazehawk.com/tester/signup/ )

Good point, we'll add some example links to the site. In the meantime:

http://www.gazehawk.com/img/techcrunch-heatmap.png

http://www.gazehawk.com/img/wakemate-heatmap.png

We pay $2 for trying right now, $4 for actually completing, and $5 if you give us some incredibly insightful feedback or bug report. We may decrease this in the future, but we need to figure out our supply/demand curve first.

The next step is to reorganize the ads on the page dynamically so you can't avoid them!

it would be extra useful if you could see the order in which people look at things

We have a nifty feature that let's you replay someone's gaze in a dot that moves around the screen. In the near future we'll have more quantitative things like how gaze order and fixation time.

awesome, once you get enough data like that you can do real cool analytics like "what color draws the most attention", "do drop shadows make it easier to focus on sections", etc... it would make for a great blog, something along the lines of http://blog.okcupid.com/

Deleted comment

TechCrunch was a fairly predictable example because blogs all tend to have the same format, but landing pages have varying flows and it's much harder to predict.

Also with TechCrunch, I wouldn't have guessed that people would look at the top right ad. I think the fact that it's in line with the top article bar leads people to it.

I would try your service in a heartbeat, but the pricing seems quite high. There's a big disconnect between charging $49 per user and paying them $5- couldn't you significantly lower pricing and still be profitable?

There's a significant overhead to eye tracking in terms of computation, so our costs aren't just paying testers. We wanted to price ourselves competitively compared to sites like UserTesting.com, but of course pricing was/is something we've discussed a lot.

I did eye tracking in college. I was told the usability lab cost in the millions, and the eye tracking stuff was a sizable portion of that.

Granted, this was many years ago, but the bottom line is still true: $50 is cheap cheap cheap compared to your other options.

If you can add where a user looks first, that'll be great. Something like order of gazes on a page?

Reply on news.ycombinator.com