Figure-ground segmentation improves handled object recognition in egocentric video | IEEE Conference Publication | IEEE Xplore