Vision-based Positioning for UAVs

Precise positioning is a prerequisite for airborne high-quality magnetic data surveying. Modern airborne magnetic surveying platforms are usually composed of two parts: an aerial vehicle and a suspended magnetometer bird. This configuration poses a challenge for georeferencing of data since the sensitivity of magnetic sensors prohibits the placement of electronic positioning sensors directly on the bird.

This work focuses on developing a vision-based positioning solution that employs computer vision and sensor fusion techniques. The solution relies on a camera and GNSS/IMU mounted on the UAV, as well as several fiducial markers on the magnetometer bird, which is a passive solution and generates negligible magnetic interference. The proposed solution comprises three primary modules, including a spatial and temporal alignment module, a fiducial detection module, and a sensor fusion module. The spatial and temporal alignment module applies a two-step calibration to estimate the parameters for synchronizing and aligning measurements from multiple sensors. The fiducial detection module processes the input video to extract regions of interest and further detects elliptical fiducials. After calculating the relative position from the detected fiducials, the sensor fusion module fuses the relative position with the global position and attitude of the UAV to determine the magnetometer bird’s global position.

We have evaluated the developed vision-based positioning solution by real flight experiments. Evaluation results show that it achieves centimeter-level positioning accuracy for a single axis and its overall 3D positioning accuracy is at the decimeter level, which comes close to the performance of the reference system. In addition, we have also contributed to pose estimation in heterogeneous space, where a series of absolute and relative pose estimation algorithms are proposed. Compared to state-of-the-art methods, the proposed algorithms demonstrate superior performance in terms of accuracy and robustness.