Many computer vision applications such as augmented reality require head pose estimation. As far as the implementation of head pose estimation on relatively resource limited mobile platforms is concerned, it is computationally challenging to meet the real-time constraints while maintaining reasonable head pose estimation accuracy. The discussed head pose estimation approach in this paper is an attempt to meet this objective. The approach consists of the following components: Viola-Jones face detection, color-based tracking using an online calibration procedure, and head pose estimation using Hu moment features and Fisher linear discriminant. Experimental results running on an actual mobile device are reported exhibiting both the real-time and accuracy aspects of the developed approach.
