Stereo-Based Head Pose Tracking with Motion Compensation Model

We present a stereo based head tracking algorithm with motion error compensation model. We firstly reconstruct 3D head model by coupling feature tracking with the epipolar constraint from the stereo image pair. Based on the reconstructed 3D model, we can obtain the coarse 3D pose of the head at this point. We then analyze the errors in the estimation of the motion parameters and introduce motion compensation model to improve the accuracy in the head pose tracking. Coupled with the optical flow constraint equation, we can get a linear equation for the unknown parameters in motion compensation model, which can be solved with low complexity. Our method is able to track all the six degrees of freedom of the rigid part of head motions, even in the presence of partial occlusions, and/or dramatic facial expression changes, and promises useful applications in human-computer interaction and eye-gaze correction for video conferencing