Multi-scale Visual Feature Association Based Long-Term UAV Positioning Algorithm

In order to solve the long-term drift of positioning in UAVs, we propose a multi-scale visual feature association algorithm, and applied in SLAM systems. Through the upward fusion mechanism, we fuse the image features extracted by the 30-degree angle for FOV and 120-degree angle for FOV, making full use of the ability to perceive global features in the wide FOV and the ability to perceive environmental details in the small FOV. Combined with IMU Sensors, we effectively suppress the positioning drift and improve the accuracy of the SLAM system. Using OSG to render the 3D real scene of Hong Kong, we obtain 3 trajectories in the environment including residential areas, city centers, bay highways, coastlines and mountains to test the algorithm. The experimental results show that the positioning error of the multi-scale visual feature association algorithm does not exceed 0.01% under extreme conditions such as the UAV flight trajectory length surpassing 10,000 m, sharp turn, and high speed, which proves that the algorithm in this paper can ensure the measurement accuracy of the positioning and effectively suppress drift of positioning, and has strong robustness.