Limbustrack: Stable eye-tracking in imperfect light conditions

Eye tracking research appears to have reached critical mass in terms of technological advances as well as sustained theoretical, applied, and even mainstream interest. There is now a wealth of interesting empirical results, performed by researchers in numerous domains, e.g., Psychology, Industrial Engineering, Marketing, and Computer Science [Duchowski 2003], to name a few. Most studies, however, are often conducted in the laboratory under various environmental constraints. Meanwhile, relatively little work has been done to collect eye movement data during performance of natural tasks, i.e., outside the lab, not to mention outside the building. Indeed, due to reliance on Infra-Red (IR) illumination of the eyeball for eye tracking precision purposes it is often difficult to track the eyes in the presence of sunlight.; Some work on collecting eye movements during dynamic activities exists, such as driving, basketball foul shooting, golf putting, table tennis, baseball, gymnastics and other situations [Rayner 1998]. However, such results come at a price, literally and figuratively. Wearable eye trackers are scarce, expensive, and/or uncomfortable (due to clumsy or heavy headgear or heavy backpacks containing laptops, cameras, etc.). We are aware of only one serious effort at development of a cheap, accurate, wearable eye tracker: the open source openEyes project [Li et al. 2006], motivated by the wearable eye tracker developed by Pelz et al. [2000]. The openEyes wearable eye tracker has matured to allow operation in the visible spectrum [Li and Parkhurst 2006]. However, its method of ocular feature detection is such that it is prone to failure in variable lighting conditions. To address this deficiency, we have developed a cheap (< {dollar}1,000) wearable eye tracker that is as comfortable to wear as a pair of plastic safety glasses. At the heart of our development are novel techniques that allow operation under variable illumination. Our wearable eye tracker provides acceptable accuracy (about 2° visual angle) at suitably fast processing speed (close to video frame rates) allowing eye movement analysis during natural tasks.