Simultaneous Localization and Mapping of high-altitude UAVs

To enhance the state estimation precision of Unmanned Aerial Vehicles (UAVs) during high-speed flight at elevated altitudes, we propose an innovative visual-inertial mileage calculation method that synergistically integrates GPS and visual attitude data for global pose refinement. The process commences with leveraging an optimized FAST corner detection algorithm to extract feature points from pre-processed images, thereby ensuring both a satisfactory extraction efficiency and an improvement in the quality of road marking point identification. Subsequently, forward and backward frame feature points are tracked and matched using a sparse optical flow technique. This step guarantees the reliability of matching point pairs by implementing dual-mode positive and negative optical flow tracking. Following this, the integration of GPS and visual rotational data is employed to rectify the local pose of the visual inertial odometer. To achieve a more stable transformation relationship, the fusion factor map is augmented with transformation factors between the local coordinate system and the global coordinate system, thus enabling the realization of global pose correction for the output results of our enhanced visual-inertial odometer. Experimental findings demonstrate that the proposed SLAM algorithm which combines GPS and visual attitude information has significantly improved positioning accuracy. Specifically, it reduces the average positioning error by 25.44% and decreases the root mean square error by 16.31%. As compared to the original algorithm, the UAV's positioning performance in real-world high-altitude environments exhibits increased robustness when utilizing the SLAM algorithm presented in this paper.