• Skin Color Detection

  • Face Detection

  • Gabor Wavelet Transform

  • The pose model with the highest similarity rank wins

  • Video-based Head Pose Recognition and 3D model matching

In my bachelor thesis I propose a method for estimating the Head Pose from Grayscale Images. With the method, we can associate statistic texture models with discrete head poses. The learned models are based on Gabor graphs, occurence probabilities and statistical textures of facial features. 93 pose-models are built and used to recognize head poses from 0°.. ±90° (yaw) and 0°.. ±45° (pitch).

An approach to extract and normalize the human face from Color Video Sequences is presented as well. It exploits Bayesian Skin color detection in CIE-lab color space; Image thresholding; Region labeling; and Viola-Jones Face Detection for frontal- & side-views of the face. Face normalization is based upon anthropometrical findings.

read B.Sc. thesis (German; ∼22 MB)

"Dynamische Kopfposenerkennung anhand statistischer Gesichtsmerkmale"