Aditya Khosla was a Facebook graduate fellow.
By Dileep Thekkethil
An Indian origin researcher-led team at the Massachusetts Institute of Technology (MIT) has found an effective eye tracking mobile application using ‘GazeCapture’ technology; the first large-scale dataset for eye tracking.
Scientists have been long determined to understand the visual sense of people who direct their eyes to one particular visual scene, neglecting others. The results of these studies are still being used for psychological experiments and also for marketing. But, such a research is highly expensive as it needs specialized equipment with heavy price tag.
Aditya Khosla, the lead researcher says that the newly developed application gives the power to the hands of everyone who wishes to track the eye without having to invest in expensive hardware other than a mobile phone or tablet.
According to the researchers, their large scale dataset called GazeCapture consists of almost 2.5M frames collected from over 1450 people and this helps the application to accurately track the eye without the need for any additional sensors other than the selfie camera.
The research claims that the iOS application called iTracker is superior to all other existing eye-tracking systems as it achieved an accurate reading, significantly reducing the chances of error.
The research paper says: “Our model achieves a prediction error of 1.71cm and 2.53cm without calibration on mobile phones and tablets respectively.”
“Since there are no applications, there’s no incentive for people to buy the devices. We thought we should break this circle and try to make an eye tracker that works on a single mobile device, using just your front-facing camera,” explained Aditya Khosla.
One of the biggest advantages of the research conducted by Khosla and his team was the amount of data they collected and worked with. The dataset of 1450 people gave the new study an edge over previous studies that had data of only 50 people.
To assemble data sets, “most other groups tend to call people into the lab,” Khosla says. “It’s really hard to scale that up. Calling 50 people in itself is already a fairly tedious process. But we realized we could do this through crowdsourcing.”
In the paper, the researchers report an initial round of experiments, using training data drawn from 800 mobile-device users.
For testing the breakthrough eye-tracking research of 2016, the team developed a simple application that can be installed on an Apple devices supporting iOS. The application flashes a small dot on the screen to attract the attention of the user and the dot, for a brief moment, is replaced with words “R” or “L”, asking the user to tap either the right or left side of the screen. If correctly executed the tap ensures that the user has shifted the gaze to the intended location. The whole process happens under the continuous monitoring of the camera that takes images of the users face.
The researchers recruited application users through Amazon’s Mechanical Turk crowdsourcing site and paid them a small fee for each successfully executed tap. The data set contains, on average, 1,600 images for each user.
This is not the first time that Khosla is teaming up with a futuristic research team. Earlier, his research on increasing number of photos being viewed and shared on the internet, what people remember, had fetched him 2013-2014 Facebook Graduate Fellowship.
Khosla completed his M.S. in computer science at Stanford University with the Stanford Vision Lab and B.S. with honors in computer science, electrical engineering, and economics at the California Institute of Technology.
The researchers will explain the new findings in the eye-tracking system in a paper that will be presented at the ‘Computer Vision and Pattern Recognition’ conference in Las Vegas on June 28.