Coarse Head Pose Estimation using Image Abstraction

Abstract

We present an algorithm to estimate the pose of a human head from a single image. It builds on the fact that only a limited set of cues are required to estimate human head pose and that most images contain far too many details than what are required for this task. Thus, non-photorealistic rendering is first used to eliminate irrelevant details from the picture and accentuate facial features critical to estimating head pose. The maximum likelihood pose range is then estimated by training a classifier on scaled down abstracted images. This algorithm covers a wide range of head orientations, can be used at various image resolutions, does not need personalized initialization, and is also relatively insensitive to illumination. Moreover, the facts that it performs competitively when compared with other state of the art methods and that it is fast enough to be used in real time systems make it a promising method for coarse head pose estimation.