Model-Free Head Pose Estimation Based on Shape Factorisation and Particle Filtering

Head pose estimation is essential for several applications and is particularly required for head pose-free eye-gaze tracking where estimation of head rotation permits free head movement during tracking. While the literature is broad, the accuracy of recent vision-based head pose estimation methods is contingent upon the availability of training data or accurate initialisation and tracking of specific facial landmarks. In this paper, we propose a method to estimate the head pose in realtime from the trajectories of a set of feature points spread randomly over the face region, without requiring a training phase or model-fitting of specific facial features. Conversely, without seeking specific facial landmarks, our method exploits the sparse 3-dimensional shape of the surface of interest, recovered via shape and motion factorisation, in combination with particle filtering to correct mistracked feature points and improve upon an initial estimation of the 3-dimensional shape during tracking. In comparison with two additional methods, quantitative results obtained through our modeland landmark-free method yield a reduction in the head pose estimation error for a wide range of head rotation angles.