Dissociating rigid and articulated motion for hand tracking

In this paper we present a novel template tracking method for articulated objects, specifically hands. The template is defined as a level set function and our goal is to distinguish between the rigid motion and the nonlinear movement of its articulations. For the rigid motion estimation, at each frame, we use a combination of optical flow and Monte Carlo sampling over the affine Lie group, Aff(2). In order to cope with the articulated parts we evolve the template using a region based active contour. Finally, we show that the use of Monte Carlo methods over Aff(2) allows the estimation of the rigid motion avoiding the noise introduced by the articulated parts.