Multi-UAV Collaborative Visual Localization with Improved Initialization and Co-view Keyframe Selection Strategy

Compared to a single unmanned aerial vehicle (UAV), multiple UAVs have significantly enhanced environmental awareness and improved fault tolerance in mission execution, where self-localization is a crucial capability for situation awareness in hazardous environments with denied or intermittent global navigation satellite system (GNSS) signals. In this paper, a multi-UAV collaborative visual localization method is improved. First, an improved multi-UAV initialization scheme is introduced which combines the Oriented FAST and Rotated BRIEF (ORB) feature and AprilTag to obtain a more accurate initial pose estimation. Second, a multi-UAV localization method is proposed based on a co-view keyframe selection strategy, by which the loop closure detection between UAVs can be implemented efficiently in real-time while sufficient localization accuracy can be obtained. Finally, the effectiveness of the proposed method is demonstrated via experiments compared with the open-source algorithm and dataset. The results show that the accuracy and robustness of the method in this paper are improved compared to the existing method.