A combined pose, object, and feature model for action understanding | IEEE Conference Publication | IEEE Xplore