Fast multi-view face tracking with pose estimation

In this paper, a fast and an effective multi-view face tracking algorithm with head pose estimation is introduced. For modeling the face pose we employ a tree of boosted classifiers built using either Haar-like filters or Gauss filters. A first classifier extracts faces of any pose from the background. Then more specific classifiers discriminate between different poses. The tree of classifiers is trained by hierarchically sub-sampling the pose space. Finally, Condensation algorithm is used for tracking the faces. Experiments show large improvements in terms of detection rate and processing speed compared to state-of-the-art algorithms.