Detail of the student project

List
Topic:Gaze direction estimation
Department:Katedra kybernetiky
Supervisor:Ing. Jan Čech Ph.D.
Announce as:DP,BP,SOP,PRO
Description:gaze

Estimating the gaze direction, i.e. estimating where a person is looking from a single monocular camera is a challenging problem with potentially large applicability. For some applications it could replace expensive eye-trackers or other intrusive 'wearable' devices. In theory, not only the direction, but also the depth of the target view can be estimated, when enough resolution is available to get the eye vergence [1].

The applications include psycho-social studies, marketing applications (e.g. contactless interaction with advertising panels), alternative mouse pointer control for disabled people, vigilance monitoring for driver assistance systems, etc. If the work is successful, there is a potential for commercialization.

Multiple approaches exist in the literature to solve this problem. A good survey can be found in e.g. [2], nice results are obtained with a help of a range sensor [3]. We will provide a code for detecting facial landmarks (nose tip, eye corners, mouth corners) in images and estimating a pose (position and orientation) of the face with respect to the camera [4]. A candidate will investigate a fusion of the head pose with the eye vergence estimation and will design a ground-truth experiment in order to evaluate the method accuracy, and to study impact of subject distance, pose, and camera resolution.

References

[1] Zakia Hammal, Corentin Massot, Guillermo Bedoya, and Alice Caplier. Eyes Segmentation Applied to Gaze Direction and Vigilance Estimation. In ICAPR 2005.
[2] Sileye O. Ba, and Jean-Marc Odobez. Multiperson Visual Focus of Attention from Head Pose and Meeting Contextual Cues. IEEE Trans. on PAMI, 33(1), 2011.
[3] Kenneth Alberto Funes Mora and Jean-Marc Odobez. Gaze Estimation from Multimodal Kinect Data. In Face and Gesture, and Kinnect CVPR Workshop. 2012.
[4] Jan Cech, Vojtech Franc, Jiri Matas. A 3D Approach to Facial Landmarks: Detection, Refinement, and Tracking. In Proc. ICPR, 2014.
Realization form:software, technical report
Date:13.05.2014
Responsible person: Petr Pošík