Robust real-time eye detection and tracking for rotated facial images under complex conditions

Abstract

In this paper, an integrated eye tracker is proposed, which can robust detect and track eyes under variable rotation angle of facial images in real time. In addition, the system is able to handle scaling, illumination changes and to detect human eyes with different distance and poses to cameras. Zernike Moments (ZM) is used for extracting the eye's rotation invariant characteristics and Support Vector Machine (SVM) is used for classifying the eye/non-eye patterns. Firstly, a face detector is used to locate face in the whole image with Haar face detection. Secondly, the innovative Template Matching (TM) is applied to detect eyes. The image is supposed to deflect, if the result of detection fails to either of two stages above. Thirdly, the Zernike Moments and Support Vector Machine (SVM) are applied to the specific area of this image by expanding search region of consecutive frame. Finally, the precise eye position is decided by the new tracker. Results from an extensive experiment show the robustness of the proposed system.