A new on-board UAV pose estimation system based on monocular camera

In visual navigation, utilizing a monocular camera to estimate the relative localization of a robot in a known environment is very common and effective. In this paper, a new method is proposed to estimate the 6D relative pose of an unmanned aerial vehicle (UAV) in a known environment. In the estimation system we assume that the landmarks locate at known positions. This method consists of three main steps. Firstly, initializing the correspondence search between landmarks and image detections with the particle swarm optimization algorithm. Secondly, applying the Kalman Filter to track and predict the pose of the UAV in the space and utilizing the predicted results to determine the correspondences between the 3D reference points and their 2D projections, excluding the negligent errors in the measurement data by Grubbs Criterion. Finally, a fast solution is employed to estimate the pose of the UAV by solving the perspective-npoint (PnP) problem. Simulation experiments demonstrate that the method we proposed outperforms state-of-the-art approaches in some aspects when the number of landmarks is larger.