In her doctoral studies, a researcher at the University of Eastern Finland has created a combined spectral video/spectral image database — the Spectral Eye Video Database, SPEED; the first of its kind.
“The Spectral Eye Video Database is a database of 51-channel spectral images of the human eye and seven-channel spectral videos of the human eye while it is performing an observational task,” says Ana Gebejes, who presented the findings in her doctoral dissertation “Spectral video: application in human eye analysis and tracking.”
In this way, the new database combines spectral acquisition and eye tracking. In fact, Gebejes says SPEED was inspired by eye tracking. Traditional eye tracking fails in harsh conditions, and the University of Eastern Finland researcher aimed to find a solution. “For this reason, some of those harsh conditions were incorporated in the database, such as reflections, shadows, makeup, extreme eye positions and eyewear.”
Gebejes modified a conventional spectral acquisition setup so that the collected data can allow for detecting the eye, tracking it and interpreting the gaze. “Spectral acquisition allows for collecting detailed spectral signatures in the eye region that are otherwise lost in monochromatic or RGB acquisition,” she explains.
SPEED features data taken from 30 voluntary subjects and comprises a total of 180 seven-channel spectral videos taken under two illumination conditions (daylight and office light) and four eye-conditions (standard, eye glasses, reflections and sunglasses). In addition, 30 fifty-one-channel spectral images were taken as well to serve as detailed spectral ground truth.
The motivation behind SPEED
This combined spectral video/spectral image database is indeed the first of its kind. Gebejes explains what motivated her to launch the project and why the database is important: “I was working with a spectral video device for about two years while completing my doctoral studies. And during this time, I took a course about eye-tracking,” she says. “It was a lot of fun to learn about the eye-tracking technology, and the more I deepened my knowledge about it, the more I started noticing that there are certain limitations and uncertainties in the field. That’s when I had my ‘aha’ moment.”
In discussing her dissertation with her professor, Gebejes realized she may be able to apply her spectral video device to solve some of those challenges through a spectral approach. “Just to give an example, in the eye region there are at least six spectrally very different features: pupil, iris, sclera, blood vessels, skin and hair,” she notes and further elaborates that in traditional monochromatic or RGB imaging the benefit of these spectral differences gets lost due to the low spectral resolution (one or three-channels respectively). “In SPEED, there is seven- and fifty-one-channel data that revels additional, spectrally dependent information that has not yet been seen or exploited in eye tracking. It opens up new possibilities, for example, multi-feature tracking, illumination-independent and spectrum-based tracking and so on...”
Gaining more information from light
“We can now extract more information from light than what we could before in the field of eye tracking,” says the expert. “When traditional detection and tracking would fail due to the spectral limitations of monochrome or RGB capture we can now exploit spectral information and create spectrally enhanced solutions.”
SPEED was originally inspired by and created for eye tracking. Nevertheless, its creator believes her SPectral Eye vidEo Database has the potential to be applied in many other eye-related areas of research — from medicine to biometrics to eye- or vision-based studies. “It can serve as a platform for creating novel spectral-data-based methodologies for imaging, training, analysis and interpretation of eye video,” Gebejes says.
Impact on future eye-tracking technologies
The spectral imaging specialist notes that SPEED provides valuable data for the creation of new spectral-data-based enhanced devices, software and algorithms. “As we now have more detailed spectra, we can exploit those to overcome difficulties, especially in tracking under harsh conditions,” she says.
“There is about 300 GB of data in front of us and a lot of ideas how to continue our research with it,” says the researcher. “I am very excited to dig into the work because every time I work with SPEED, I find something new and fascinating.”
To Gebejes, understanding the spectral stability of features in the eye region reveals interesting details, and that is where she is going to continue to focus her work.
Written by Sandra Henderson, Research Editor, Novus Light Technologies Today