LifeCam VX-1000 Eyetracking

Towards easy-to-use and robust low-cost eyetracking systems. 30Hz version of 30 September 2008.

Hardware: 1 x Microsoft LifeCam VX-1000, IR-blocking filter removed; 2 x Lightbulb 15 Watt.

Software: Homebrew using Visual Studio 2008, C++, MFC, and Intel's Open Computer Vision Library (OpenCV).

Accuracy test: One of 48 screen 'tiles' is randomly selected and an on-screen gaze target appears at its center. The user gazes at this target. In the visualization a blue dot denotes the 'raw' gaze direction of the left or right eye. The eye symbol is the low-pass filtered arithmetic mean of these two values. This is a good estimator for the current point of regard. Whenever the estimate does not hit the selected tile, the tile is colored red. When the estimate is within the boundaries of the selected tile, the latter is colored green. Finally, the relative tile hit time is computed as a test score for the overall accuracy.

Thank you very much for your attention!

Category:

People & Blogs

Tags:

• LifeCam
• eyetracking
• computer
• vision
• low-cost
• C++
• MFC
• OpenCV

[tomasmk]

is it possible to do this with java?

[piechulla1966]

@tomasmk

Yes, You can use JavaCV - it contains Java wrapper classes for the full C/C++ API of OpenCV 2.3.0, also works with Android. Personally, I never tried JavaCV. I only know that it exists.

[tomasmk]

@piechulla1966 thanks!

[piechulla1966]

@tomasmk

This is a good question. I bought a Thorlabs DCC1545M (nice camera) in 2010, but I was too busy to try it out. Shall play around with this interesting device next week. Point Grey exhibited a USB3 camera prototype at the Vision 2010 fair (Stuttgart, Germany). In the near future, USB3 could give us high resolution at high frame rates -- exactly what we need :-)

[piechulla1966]

@tomasmk

Windows boxes (XP/Vista), MS Visual Studio (2008/2010), OpenCV (1.0/1.1/2.0/2.1/2.2) and some years of experience in tweaking high-end eyetracking systems (in order to be able to use these systems for "cognitive ergonomics" aka "man machine interaction" aka "human factors").

[tomasmk]

@piechulla1966 thanks, will do. when will you finish your eye tracking program?

[tomasmk]

@piechulla1966 thanks, will do. what did you use for this program?

[piechulla1966]

@tomasmk

Active Contour Models (Snakes) seem to be helpful. I never tried it. Zafer Savas gives an introduction in his MSc thesis (just google for his name). Pages 64-68 explain his Snakes approach to eyelid tracking.

Cheers,

walter

[tomasmk]

@piechulla1966 yes you are right, but my proyect is intended to detect only when a driver actually falls asleep or closes his eyes for more than 2 seconds. A couple of weeks ago I had a near-death experience in wich i kind of fell asleep for 3 or 4 seconds while I was driving a car with my family in it. Fortunately I woke up when the car was a couple of inches from crashing into a truck. A simple alarm activated by my eyes being closed for too long would have eliminated any danger.

[piechulla1966]

@tomasmk

PERCLOS or similar indicators of drowsiness are non-trivial. A webcam is too slow to detect eyeblinks and estimate eyeblink duration. Go for a good camera that gives you framerates of 100 Hz or faster.

Cheers,

walter