UAV Autonomous Landing Pose Estimation Using Monocular Vision Based on Cooperative Identification and Scene Reconstruction

Nowadays UAVs are often employed in a weak or GPS signals unavailable environment which requires high demand for the UAV landing autonomously. In this paper, propose a monocular vision UAV autonomous landing pose estimation method based on cooperative identification and scene reconstruction. A rotating target detection algorithm specific to aerial images is used to identify and locate the target with the apron in aerial images. When the UAV lands to a height where the detail information of the cooperative identification on the apron can be extracted, firstly, the key points are extracted from the single frame images acquired by the airborne monocular camera and matched with the key points saved in the airborne database for geometric verification to filter out the wrong matching relationships. Then, the 3D coordinates of feature points saved in the onboard database are used to obtain the 2D-3D matching relationship and perform co-visual relationship screening to obtain stable matching relationships. Finally, the PnP problem is solved by BA optimization method, and the position and yaw angle of the UAV relative to the mobile apron are calculated according to the similar transformation matrix saved in the airborne database. The experiment results indicate that the proposed method improves the accuracy of UAV pose estimation and can be adopted as an alternate UAV autonomous landing technology in a narrow mobile apron.