We're currently learning this: http://www.mathworks.com/help/vision/examples/face-detection-and-tracking-using-the-klt-algorithm.html

In short, we're trying to detect face and its features, even if the person shook his/her head.

I also happened to know another algorithm called Viola-Jones, but as long as I know, it can only detect if the face is straight ahead to the camera and won't work when the person shook their head.