Nava Chitrik and Yuliy Schwartzburg have in partial fulfillment of their Senior Design Project Requirements constructed a low-cost approach for remote eye tracking at the Cooper Union for the Advancement of Science and Art, Electrical Engineering Department.
"The line of a person's gaze is known to have many important applications in artificial intelligence (AI) and video conferencing but determining where a user is looking is still a very challenging problem. Traditionally, gaze trackers have been implemented with devices worn around the user's head, but more recent advances in the field use unobtrusive methods, i.e. an external video camera, to obtain information about where a person is looking. We have developed a simplified gaze tracking system using a single camera and a single point source mounted compactly in the view of the user, a large simplification over previous methods which have used a plurality of each. Furthermore, our algorithms are robust enough to allow head motion and our image processing functions are designed to extract data even from low-resolution or noisy video streams. Our system also has the computational advantage of working with very small image sizes, reducing the amount of resources needed for gaze tracking, freeing them up for applications that might utilize this information.
To reiterate: The main differences between this implementation and similar implementations are that this system uses a histogram method as opposed to edge detection to work with very low resolution video extremely quickly. However, it requires an infrared camera and infrared LED's. (Which can be purchased for less than 25 dollars online.)"
View on YouTube
"The line of a person's gaze is known to have many important applications in artificial intelligence (AI) and video conferencing but determining where a user is looking is still a very challenging problem. Traditionally, gaze trackers have been implemented with devices worn around the user's head, but more recent advances in the field use unobtrusive methods, i.e. an external video camera, to obtain information about where a person is looking. We have developed a simplified gaze tracking system using a single camera and a single point source mounted compactly in the view of the user, a large simplification over previous methods which have used a plurality of each. Furthermore, our algorithms are robust enough to allow head motion and our image processing functions are designed to extract data even from low-resolution or noisy video streams. Our system also has the computational advantage of working with very small image sizes, reducing the amount of resources needed for gaze tracking, freeing them up for applications that might utilize this information.
To reiterate: The main differences between this implementation and similar implementations are that this system uses a histogram method as opposed to edge detection to work with very low resolution video extremely quickly. However, it requires an infrared camera and infrared LED's. (Which can be purchased for less than 25 dollars online.)"
View on YouTube