In this paper, we describe a method for estimating 3D pose of a human head from a 2D monocular web-cam image. We propose a novel system based on existing computer vision techniques for real-time head pose estimation, which can start and recover from failure automatically without any previous knowledge of the user's appearance or location and keeping the user free of any devices or wires. The system is robust to large pose and facial variations and to partial occlusions. With this setting a widespread use on different machines e.g. computers or laptops is guaranteed. The head pose will be estimated directly from the web-cam image appearance, leaving the user completely free of any artificial markers or special glasses. Furthermore, this system estimates the orientation in the 6 degrees freedom with an uncalibrated monocular camera. The system tracks a person's head pose at a reasonable distance without any camera calibration. Finally, the experimentation on the proposed systems shows that it is accurate, fast (23 fps, 43 ms) and is flexible to use in cases where classic approaches would fail.     «

In this paper, we describe a method for estimating 3D pose of a human head from a 2D monocular web-cam image. We propose a novel system based on existing computer vision techniques for real-time head pose estimation, which can start and recover from failure automatically without any previous knowledge of the user's appearance or location and keeping the user free of any devices or wires. The system is robust to large pose and facial variations and to partial occlusions. With this setting a wides...     »