Eyelids and face tracking in real-time

Abstract

Tracing and tracking facial features in a precise manner are crucial tasks for Human Computer Interaction, facial expression recognition, and image retrieval. Eyelids tracking is an important and hard subject to evaluate human emotions due to the high expressivity of the eyes and its faster movement. For Active Appearance Models (AAM) is a challenge, since related frameworks have showed a good performance using edge detectors, color information, and thresholding techniques all of them depending on the quality of the image. We built two appearance-based models (ABM) to track simultaneously eyelids tracking with 3D head pose, eyebrows and lips in monocular video sequences. This paper has two main contributions. Firstly, we show that by adopting a non-occluded facial texture model in the eyes region, 3D head pose parameters can be obtained in an accurate and stable manner. Secondly, unlike previous approaches regarding eyelids tracking, we prove that the Online Appearance Models (OAMs) can be used for eyelids tracking without color information, eyes feature extraction, or edge detectors. Experiments in real videos show the feasibility and usefulness of this approach, down-weighting time and memory consumption, and improving the accuracy.