A robust head pose tracking system based on multiple cameras

Tracking human head pose has many applications such as building input devices for game players and for the disabled. Traditional vision methods detect the image of the head to deduce its pose. Our approach is the reverse, we attach cameras to the head so head pose tracking becomes camera pose tracking. The main problem with camera centered vision based pose tracking is that, if one camera is used, translation and rotation will produce very similar effects on the images hence the pose cannot be correctly detected; this problem is called the translation/rotation ambiguity problem. The novelty of our approach is that we attach multiple cameras to the head to tackle the ambiguity problem and obtain very accurate pose result. As the costs, sizes and weights of cameras are reducing quickly, this is in fact a very practical approach for making a head pose input device. Experiments on simulation and real data have shown the effectiveness of the method.