Research Goal | Eye Tracking | Gaze Tracking | Face Tracking |Facial Feature Tracking |Facial Motion Recovery | Facial Expression Recogntion

Real-Time Facial Feature Tracking Under Significant Facial Expressions and Various Face Orientations

 

 

 

Facial features, such as eyes, eyebrows, nose and mouth, and their spatial arrangement, are important for the facial interpretation tasks based on face images, such as face recognition, facial expression analysis and face animation. Therefore, locating these facial features in a face image accurately is a crucial step for these tasks to perform well. However, in reality, the appearance of the facial features in the images varies significantly among different individuals. Even, for a specific person, the appearance of the facial features is easily affected by the lighting conditions, face orientations and facial expressions, etc. Therefore, accurate facial feature detection and tracking still remains a very challenging task, especially under different illuminations, face orientations and facial expressions, etc.

In our research, we proposed an effective approach to detect and track twenty-eight facial features from the face images with different facial expressions under various face orientations in real time. The improvements in facial feature detection and tracking accuracy are resulted from: (1) combination of the Kalman filtering with the eye positions to constrain the facial feature locations; (2) the use of pyramidal Gabor wavelets for efficient facial feature representation; (3) dynamic and accurate model updating for each facial feature to eliminate any error accumulation; (4)imposing the global geometry constraints to eliminate any geometrical violations. By these combinations, the accuracy of the facial feature tracking reaches a practical acceptable level. Subsequently, the extracted spatio-temporal relationships among the facial features can be used to conduct the facial expression classification successfully.

 

Facial Feature and Pose Tracking based on 3D Deformable Models

 

 

 

In our recent work, we compared our method with 3D tracking method based on the Candide-3 face model and particle filter. The 3D deformable face model can be manually initialized or be automatically initialized at the first frame with 28 detected facial feature points detected by our 2D tracker. Demo 5 gives the 3D tracking result.  Demo 6 shows the comparison of 2D and 3D tracker and indicates that the 3D tracker can also reliably track 28 salient facial feature points, and provide more other facial feature points like those on the cheek. Even though currently we only take into account the rigid motion, the Candide-3 face model also allows non-rigid motion and the Action Unit (AU) parameters of this model can be directly applied for expression analysis. It can be seen from Demo 6 that facial feature tracking and pose estimation result of the 3D tracker are not as smooth as those output by our 2D tracker, this can be explained by the fact that this 3D tracker is purely global feature based (in the 3D tracking method, the whole face will be warped with the estimated parameters of Candide-3 into geometrically-free facial patch, and then compared with a template to output the likelihood), while our 2D tracker is also based on local search.

 

We also tried directly instantiating the 3D face model with 28 facial feature points provided by the 2D tracker at each frame. The benefits of doing this include providing more facial feature points and the AU parameters. The result is shown in Demo 7Exploiting the cooperation between 2D and 3D face model would be one of our future works.

 

 

 

Publications:

(1) Yan Tong and Qiang Ji, “Automatic Eye Position Detection and Tracking under Natural Facial Movement”, Passive Eye Monitoring, Editor: Riad I. Hammoud, pp. 67-84, Springer 2007.
(2) Yan Tong, Yang Wang, Zhiwei Zhu, and Qiang Ji, “Robust Facial Feature Tracking under Varying Face Pose and Facial Expression”, Pattern Recognition, Vol. 40, No. 11, pp. 3195-3208, November 2007.
(3) Zhiwei Zhu and Qiang Ji, Robust Pose Invariant Facial Feature Detection and Tracking in Real-Time, the 18th International Conference on Pattern Recognition (ICPR), Hongkong, August, 2006.
(4) Yan Tong, Yang Wang, Zhiwei Zhu, and Qiang Ji, “Facial Feature Tracking using a Multi-State Hierarchical Shape Model under Varying Face Pose and Facial Expression”, the 18th International Conference on Pattern Recognition (ICPR), Hongkong, August, 2006.
(5) Yan Tong and Qiang Ji, “Multiview Facial Feature Tracking with a Multi-modal Probabilistic Model”, the 18th International Conference on Pattern Recognition (ICPR), Hongkong, August, 2006.

Demos:

 

 

 

Demo.1 Real-time facial feature tracking demo (short version)

Demo.2 Real-time facial Feature tracking demo (long version)

Demo.3 Real-time facial feature tracking demo with estimated 3D head pose

 

 

 

Demo.4 Real time facial feature tracking under large pose and scale changes

 

 

Latest Demos:

Demo.5 3D tracking based on the Candide-3 face model (Initialized by 28 detected facial feature points), Bottom is the estimated pose, top left is the geometrically-free face template

Demo.6 Comparison of 2D and 3D facial feature/pose tracking (Red/Blue circles represent results of 2D/3D tracker, Bottom left/right is the estimated pose by 3D/2D tracker)

Demo.7 3D face model fitting with facial feature points provided by 2D tracker (Top right is the warped face, the bar graph at bottom right indicates the intensities of AUs)