Using iris region eigenvalues for user-independent eye-gaze distance estimation in single-camera setup

Abstract

A novel user independent distance measurement method for eye-gaze systems based on the eigenvalues of an iris is presented in this paper. The eye-gaze images are captured using only one IP-camera with the distance between the user’s face and the screen approximately 60 to 80 centimeters. Each distance is composed of 60 predefined points on the screen and the eigenvalues of the iris from a single user is used for modeling distance estimation using the exponential, linear, logarithmic, and power regressions. The test was conducted on 30 subjects who were not a part of the training process. The experimental results substantiate the effectiveness of the exponential model which yielded 96.69% accuracy with the mean absolute error (MAE) of 2.33 centimeters. The test results obtained from linear, logarithmic, and power regressions were 95.98%, 95.33%, and 96.37% accuracy with the MAE’s of 2.78, 2.68, and 2.59 centimeters, respectively. The proposed eye-gaze distance estimation method shows high accuracy with low MAE in distance estimation compared to the pixel count and gray-level intensity approaches whose accuracy is 92.31% and 96.54%.