Finding approximate head orientation in non-intrusive environment with single camera

摘要

Finding head orientation is an important issue in Human Computer Interaction (HCI) for disabled people. It also has applications ranging from drivers attention estimation to monitoring vigilance level of the audience. Head pose detection in non-intrusive or in non-highly specialized environments is, most of the times, difficult to achieve. Most of the work in this domain needs prior knowledge regarding the equipment setup or specialized hardware. We present here an appearance based technique for finding head orientation where environment consist of single camera only. No prior knowledge about environment, like distance between camera and person sitting in front of the camera, is required. Orientations are found in terms of Yaw, Pitch and Roll angles. Our method finds approximate head orientation as we have not calibrated the camera, but it still suits various applications like vigilance monitoring or HCI systems as these applications require general head rotation patterns rather than exact orientation angles.