I've made some progress. Since I'm short on time, I'll just summarize.
First Steps
I began with the intention of building an
Eye Writer 2. While the Eye Writer 2 does appear to work reasonably well as a remote tracker (provided a chin rest is used), the solution seemed to be primarily geared towards the Eye Writer software and is very Macintosh-centric. It appears that it may be used on other systems.
Course Correction
I abandoned the idea of building an Eye Writer 2 because I want to use Ogama and ITU Gaze Tracker to do some eye tracking research (for a multi-user virtual environment we are building). It was not clear that the Eye Writer 2 would interface well with ITU GT.
Prototype 1 Build
I attempted to build a remote eye tracking setup using a hacked Sony PS Eye camera with a 12mm lens and 2 external IR LED arrays, as outlined
elsewhere. While I was able to get the tracking hardware working, calibration was hit-and-miss. Binocular was less accurate than monocular.
Prototype 2 Build
I then acquired the
cheapest USB webcam I could find. I removed everything from the PCB, replaced the LEDs with 2 IR LEDs, installed a visible light filter and installed a 6mm lens. I then attached it to a pair of sunglasses from which I removed the lenses. While I didn't follow these instructions to the letter,
this pretty much sums up what I did. Good information can also be found
here (and the full article
here)
The second prototype provided greater accuracy, but the junky webcam I used has a low frame rate and the optics are noticeably inferior to those of the PS Eye. I blogged about my initial results
previously.
Lessons Learned (So Far)
The first lesson is that the devil is in the details. While you can build a cheap eye tracking system on the cheap, investing in quality gear will make a difference. Spend your money on your webcam and get some decent lenses (Peau Productions has great stuff!). Everything else can be hacked together, but optics and framerate are critical to accuracy.
The second lesson is that remote tracking with webcams is hit-and-miss, very finicky, and frankly too brittle to be useful for serious research. Head-mounted eye tracking with webcams, on the other hand, can be very accurate (see
this article for an in-depth analysis). Unless you can afford the ThorLabs cam, plan on head-mounted and save yourself time and hassle.
The third lesson is that head-mounted has its own set of challenges. The camera likes to shake. Head movements (even very slight) disrupt accuracy and throw off calibration. Finding the proper distance to sit from the monitor in order to get good tracking results is a challenge. Do bigger displays make a difference? What about projected displays? I'm investigating this.
Recommendations and Next Steps
Most of the discussions in this forum focus on head-mounted or remote setups. Both of these designs have their strengths and weaknesses. My recommendation is a hybrid setup for cheap webcams (not recommended for ThorLabs cam). In order to maintain low cost while achieving high accuracy, I propose attaching the webcam to a chin rest. This way, the webcam is not attached to the user's head, reducing shaking and movements and increasing comfort. In addition, the user's head is kept still, helping to increase accuracy. An illustration is provided below.
Next steps will be to build a DIY chin rest and attach the PS Eye camera to it. I will also add IR LEDs close to or on the PS Eye camera as opposed to using an external IR LED array. Stay tuned.
