Head pose estimation using facial-landmarks classification for children rehabilitation games

During the last decade, there has been an increasing interest in developing Head-Pose Estimation (HPE) methods for different applications. Among these, there is the possibility to track a children's head pose during rehabilitation sessions and to use such information to control a virtual avatar, so to increase the engagement and the effectiveness of the exercises. This requires the ability to perform such tracking in real-time, with high precision, and considering wide set of tracking angles. HPE methods can be generally categorised either as appearance-based or model-based methods, while, in this paper, we propose a novel, simple but effective, hybrid method for estimating the Head-Pose. It starts by detecting the face followed by detecting robust feature points on it (facial landmarks). The second part consists of applying a classification mechanism to assign facial landmarks characterising a face to a predefined range of angles representing the face orientation. The obtained results allow using the proposed approach in real-time and showed the efficiency of this approach to get significant improvement compared to the state of the art.