Abstract: A novel monocular vision method for landing of unmanned aerial vehicle (UAV) to estimate its state relative to a known two-circle planar target is proposed. The motion and projection model of the camera on-board of UAV is established. Twenty-one control points with invariance of perspective projection are generated with eight common tangent points of two coplanar circles onto the designed target. The position and orientation of the camera can be computed with the world coordinates of twenty-one control points and the corresponding image coordinates. The feature extraction of two circles and control points labeling in a 768×576 pixel size image with complex background may be completed within 9ms. Simulation test results show that the proposed vision-based state estimations are accurate to within 6cm refered to each axis of translation and 0.7 degrees refered to each axis of rotation when the distance between the camera and the landing pad is about 10m and the Gaussian white noise level is 1.5 pixels. The proposed technique is robust to noise and reliable, and it can meet the demand of the real time measurements of the position and orientation for control of UAV.

Key words: unmanned aerial vehicle, position, orientation, feature point, perspective projection