A hovering control method based on onboard monocular vision is proposed to hover a micro aerial robot autonomously, in which there is no external positioning system in indoor environments. A descriptor with four components and a multi-stage filter are used for feature tracking. Horizontal position is estimated according to monocular vision kinematics. Flight speed is estimated according to aerodynamic drag at low Reynolds number. Position and velocity informations are fused to hover the robot. Experimental results show the effectiveness of the proposed approach.