Active object recognition by view integration and reinforcement learning