Geometric approach for pose detection of moving human heads

Presents the detection of 3D pose of moving humans heads. The contribution of this work is the conjunction of an effective preprocessing method and a strong estimation Kalman technique which uses less noise sensitive 3D line observations. The procedure involves the detection of the target face in any type of background using color histograms. To outline the face we use a minimization algorithm for the determination of its tangent directions. Face 3D lines are coded in terms of motors (dual quaternions) which are used by a motor extended Kalman filter for the estimation of the 3D pose of the head through time.