Contextual Visual Object Recognition And Behavior Modeling For Human-Robot Interactivity