Pose estimation of human body part using multiple cameras

In this paper, we present a method of obtaining the approximate transformation parameter values as a starting point in estimating the pose of rigid 3D free form objects using multiple 2D images. We back project the edge silhouettes in the images, and obtain the approximate volume in the 3D space containing the object. Next, for a point selected in the volume, we hypothesize a set of points within the 3D CAD model of the object it can possibly correspond to, using the spatial extent function introduced in this paper. This is repeated for three arbitrarily chosen point in the volume. The hypothesized (match point) lists of these three points are next used to derive the pose parameter by enforcing the conditions of rigidity. Our initial experiments demonstrate the potential of this idea, and the pose parameters estimated using this method can be refined using the standard methods available in the literature.