Integrating Head Pose Tracking with Eye Tracking for Wider FOV Virtual Environment

Abstract

In virtual environment based driving experiments, eye-tracking is very helpful for quantitatively investigating driver visual behavior. However, commercial eye trackers usually have a limited maximum tracking range, e.g. ±35 degrees in horizontal direction for Tobii x50. In high fidelity applications with 135 degrees view, or even 360 degrees curved screen, there is a need to extend the range of eye tracking. This paper proposed a method to combine head pose tracking and eye tracking to achieve a large range of sightline tracking in wide field of view applications. Head poses (yaw and pitch) were estimated from head images with Multilayer Perceptrons (MLP). Head images were represented with a combination of coefficients obtained from Principal Component Analysis (PCA). With a 7times6 calibration grid, head pose estimation achieved the mean errors of yaw angle at −2.1° and pitch angle at −6.5°. The standard deviation was about 10°. The estimation process is feasible for real-time wide field of view virtual environment applications.