A UAV location method based on aerial-satellite image matching and factor graph optimization in GNSS denied aera

Abstract Global navigation satellite system (GNSS) standalone or GNSS/Inertial Navigation System integrated apparatus is the essential part of a low-altitude Unmanned aerial vehicle (UAV) to achieve reliable location information. However, GNSS is fragile and easily attacked, it is critical to develop other absolute location methods for UAVs under GNSS-denied environments. In this paper, an aerial-satellite (AS) image matching based visual location system (AS-VPS) is proposed. Two key modules ‘front-end visual odometry’ and ‘back-end absolute location’ are included in the AS-VPS. A deep learning method ‘SuperPoint + LightGlue’ is utilized for front-end consecutive aerial image matching for visual odometry and back-end AS image matching for absolute location. In view of the characteristic that the back-end absolute location is not continuous, a factor graph optimization (FGO) method is then employed to integrate the position results from the front-end and back-end modules, and final optimized results are generated. In addition, a cooperative AS-VPS system is set up using the FGO method with inter-agent ranging measurement. In the experiment, real-world flight-testing datasets are collected to assess the performance of the proposed AS-VPS method, and the experimental result demonstrates the superior performance of the proposed method without GNSS.