StereoTracking

Face Tracking in Stereo Vision

Beckman Inst. UIUC, 2001

Facial tracking is a key step for face modeling, coding and animation. We present a facial tracking method in stereo vision. We use an efficient representation of 3-D face motion, consisting of global head motion and local facial deformation. Stereo vision is utilized to measure 3-D coordinates of the face, the global and local motion parameters of the tracked face is solved respectively. From the 3-D space to the left and right image space we employ re-projection to realize 3-D tracking. Derivation details are shown in our paper ICIP'02.

3-D Face Model and its Fitting on a Face Image

Modeling the face motion space by a set of harmonics-based orthogonal bases

Calibration model and the detected corners with 3-D ground truth

Normalized cross correlation for point matching

Initialization in tracking with our stereo data

Tracking with MSR stereo data (vertically mounted a pair of cameras)

Welcome to Yu Huang's homepage!