3D motion estimation of video objects using a priori data and 2D apparent motion

This paper deals with the estimation of the three dimensional motion of a locutor appearing in a real video sequence. This is done using a new head tracking method in which the head position is modeled by means of a parametric elliptic contour fitted by minimization of a suitable objective function. The ellipse is tracked using a Kalman approach and its successive positions allow the computation of the head´s apparent motion. Finally, the three dimensional translatory movement is determined using the estimated apparent displacement and a priori information of head size