Accurate and robust head pose and -motion estimation is crucial for human behaviour analysis for two reasons. First of all, head movements convey an important part of a person’s self-expression. Secondly, the appearance of the face in any specific camera view, depends heavily on the relative head pose. Therefore, analysing facial expressions can only be done robustly when the head pose is taken into account.
Obtaining a sufficiently accurate and reliable ground truth of head pose is crucial for development and testing of methods for marker-free estimation of head pose. But also for doing research on methods for facial expression analysis that assume and rely on a prior availability of head pose information.
Marker-based optical motion capture used to rely on infrared markers, and an optical filter that blocks the visual spectrum. Unfortunately, this renders the images useless for further research on automatic human behavior understanding.
To overcome this problem, we have combined two innovations to the original approach of optical marker-based motion capture:
The result is an algorithm that works fast enough to be used in real-time applications such as human head pose estimation.
J. Lichtenauer, M. Pantic. Proceedings of the 6th IEEE Workshop on Human Computer Interaction: Real-Time Vision Aspects of Natural User Interfaces in conjunction with ICCV 2011. Barcelona, Spain, pp. 430 - 436, November 2011.