An approach to extract Natural Grasping Axes with a real 3D vision system

We address here the problem of grasping 3D objects whose geometrical envelopes are provided by a visual system. A previous article described a grasp planner based on the human grasp for virtual objects modelled with a rendering software. We made evident that, during the approaching phase, human beings shape their hands to comply with what we call the natural grasping axis, parallel to the palm and surrounded by the thumb facing the other fingers. However to estimate the grasping axes for real objects, it is crucial to extract their 3D geometry from their environment. In this paper, we propose to extend this original methodology for grasp planning by implementing an algorithm for 3D reconstruction with multi-cameras: the Voxel coloring. It consists in decomposing the scene to be reconstructed into small elements of volume projected onto the image set. A complete simulation example that illustrates the performance of the approach is given at the end