Semantic Scene Understanding for Human-Robot Interaction