Visual 3D motion estimation of UAV and landing target based on extended Kalman filter

During the last phase of auto landing an unmanned aerial vehicle (UAV) on the ship, the estimation of the 3D relative motion parameters between UAV and the landing target can be regarded as the planar 3D motion estimation between the camera mounted on the UAV and the deck. An algorithm for visual motion estimation of 3D objects based on extended Kalman filter is presented. First, the camera coordinate with the origin at the camera lens and the world coordinate are set up appealing to the principles of perspective projection. Then, the actual 3D camera motion parameters (the three Eulerian angles, transition vectors and their velocities) can be described in terms of the state equation. Furthermore, with the target corner extraction and frame matching, the observation equation is proposed to give the relationship of the feature points in the image and the state vectors. All the 3D relative motion parameters are solved by the stated extended Kalman filter (EKF) method. The presented experimental results of both synthetic data and the real image sequences show that our algorithm is effectively and robust.