3-D Gaze-Estimation Method Using a Multi-Camera-Multi-Light-Source System
Jiannan Chi,Jiahui Liu,Feng Wang,Yingkai Chi,Zeng-Guang Hou
DOI: https://doi.org/10.1109/tim.2020.3006681
IF: 5.6
2020-12-01
IEEE Transactions on Instrumentation and Measurement
Abstract:The image of the center of a 3-D circular target is not the center of its imaging ellipse due to the imaging distortion of the spatial circular targets; therefore, traditional 3-D gaze-estimation methods with the multi-camera-multi-light-source (MCMLS) systems generally replace the projection of the pupil center with the virtual image of the pupil center for 3-D gaze estimation. However, this introduces a large gaze-estimation error when the oblique angle between the optical axis of the eye and the camera optical axis is large. To eliminate the error caused by using the virtual image of the pupil center, a 3-D gaze-estimation method using an MCMLS system is developed in this study. We first estimate the cornea center and then determine the matching points of the pupil imaging ellipse using the special polar plane. After the cornea radius and the kappa angle are calibrated, the optical axis of the eye is reconstructed by the refraction planes constructed by the edge points of the pupil imaging ellipses. Thus, the 3-D gaze is estimated by the real-time calculated transformation matrix represented by the kappa angle. The feasibility and performance of the method have been analyzed by the simulations and the experiments. Since the estimation of the spatial pupil center or the construction of the refraction plane using the virtual image of the pupil center is not required, the proposed method mitigates the inherent errors caused by optical axis reconstruction in the traditional methods and simplifies the algorithm for gaze estimation, which has a practical value.
engineering, electrical & electronic,instruments & instrumentation