Pose estimation method based on artificial landmarks for the autonomous flight of underground UAVs

UAVs (Unmanned Aerial Vehicles) with small size, light weight and good maneuverability will play a significant role on the unmanned coal mining in the future, and the pose estimation is the key to the au-tonomous flight of underground UAVs.Aiming at the pose estimation of underground UAVs autonomous flight along straight or lower curved tunnel, the reflective tags with the same shape, which are deployed in pairs along both sides of the underground roadway wall in parallel, are proposed as the artificial landmarks to assist underground UAVs in pose estimation.In order to realize the autonomous flight of UAVs in the mine environment without GPS, considering the somber characteristics in underground coal mine, it is proposed that the monocular camera in the mine roadway adopts visual attention mechanism to extract the outline of the reflective tags.In order to reduce the amount of calculation, the conventional algebraic method is replaced with geometric methods.It will avoid a complicated nonlinear solution for every frame of image in the calculation process of solving the position and attitude parameters by the traditional feature point method.According to the geometrical relationship between the lens center of the monocular camera and four adjacent reflective tags as well as their projection points, taking the world coordinates and pixel coordinates of the reflective tags as input, based on the volume measurement of tetrahedral composed of any three of the four adjacent reflective tags and the lens center of the monocular camera, the coordinates of four reflective tags in the camera coordinate system are derived with a geometrical method.According to the transformation relation between the known world coordinates and the obtained camera coordinates of the reflective tags, the transformation parameters between two coordinate systems, i.e.rotation matrix and translation matrix, are obtained with the singular value decomposition (SVD) method, which provides the pose estimation of UAVs in straight or lower curved mine roadway.The experiment results along corridor to simulate straight or lower curved mine roadway show that the average location error is around 0.076 m, and the average angle error is about 1.74° with the proposed estimation algorithms for underground UAVs, and the single-frame computing time is only 0.013 s which will improve the instantaneity of pose estimation of underground UAVs.