IR Image Based Eye Gaze Estimation

Abstract

In this paper, we present a novel approach to estimate the human's eye gaze from a single infrared spectrum image of two eyes. Observing that the pupil contour is a circle, we estimate the normal direction of this circle, considered as the eye gaze, from its elliptical image. By assuming that the two gaze vectors of both eyes and the vector which is determined by the two centers of the right and left pupil are coplanar, we can calculate the both eye gaze. Compared to most existing 3-D gaze estimation algorithms, our approach does not require those of face feature information such as the eye corners. Since the focal length can be calculated by our "Coplanar Constraint", ours does not need to know the exact focal length of the camera. The extensive experiments over simulated images and real images demonstrate the feasibility and the effectiveness of our method.

A Multi-Gesture Interaction System Using a 3-D Iris Disk Model for Gaze Estimation and an Active Appearance Model for 3-D Hand Pointing

“…Alternatively, eye gaze may be determined from the pupil or iris contours [17] using ellipse fitting approaches [18], [19]. With a sufficiently large image of the eye, the iris contour and reflection of the screen off the cornea can be used to determine gaze [20].…”

“…Our first constraint can be expressed as follows: (17) We also know that the dot product of a vector with itself is its length squared; so, after multiplying out and gathering terms, we get (18) Let us now look at our other constraint. We want to find a vector going from to such that it intersects at .…”

Nonintrusive Eye Gaze Tracking using a single eye image

“…In recent years, considerable attention has been given to EGT due to its potential applications in fields such as HCI [1][2][3][4] and human cognitive studies (HCS) [5][6][7][8][9][10]. Tremendous efforts have been devoted to nonintrusive [11][12][13][14][15][16][17][18][19][20] as well as intrusive EGT techniques [21][22][23] in order to develop more accurate, faster, and easier-to-use systems. Nonintrusive methods have been much preferred over its intrusive counterpart due to the potential comfort when performing EGT.…”

“…Other nonintrusive works such as [11,12,14,18] also report high accuracy, but with similar requirements of multiple physical and technical resources, which are normally expensive to acquire. Intrusive systems, understandably has the potential of giving high accuracy due to the various devices that need to be attached to the user (e.g [21,22]).…”

Using eye gaze, head pose, and facial expression for personalized non-player character interaction

“…However, eye gaze may also be determined from pupil or iris contours [22], ellipse-fitting approaches [4], [20], the distance between the iris center and certain reference points (e.g., the eye corners) [14], [23], eye region segmentation and pixel-wise matching with 3D rendered eyeball models [21], [24], or even the reflection of the screen off the cornea [16]. Some previous work shows successful head pose estimation based on a depth-aware camera [1], multiple cameras [7], and a single camera [18], but accurate, continuous, and real-time head pose estimation remains a challenge, particular with large head rotations.…”