Vision-Aided State Estimator for Positioning UAVs

Recent developments of UAVs have sparked interest in building UAV-based magnetic surveying systems. Due to the sensitivity of magnetic sensors, direct positioning the magnetometer via traditional devices, e.g., GNSS and UWB, is not possible. However, to get obtain georeferenced data, it is necessary to localize the payload accurately.

This paper presents the sensor fusion technique used for the development of an accurate vision-based positioning solution. The sensor fusion module builds upon a square-root unscented Kalman filter to fuse the relative positioning results computed by a monocular camera with the global positioning results from a GNSS/IMU system to determine a target's global positions.

The developed method has been validated by real flight experiments. Experimental results show it can provide accurate position estimations, achieving single-axis centimeter-level positioning accuracy and decimeter-level overall 3D positioning accuracy.