LL-SLAM: Lightweight Loosely-Coupled Visual-Inertial SLAM

This paper presents LL-SLAM, a lightweight visual-inertial simultaneous localization and mapping (SLAM) system based on the loosely coupled method for autonomous flight and navigation tasks of the unmanned aerial vehicle (UAV). LL-SLAM consists of the stereo visual pose estimation module (visual module) and the EKF-based inertial pose estimation module (inertial module), which have complementary strengths. LL-SLAM integrates the poses of two modules into an accurate and robust pose estimation according to the tracking status using the loosely coupled poses integration algorithm. The system innovatively uses inertial poses to provide prior poses for visual module and uses visual poses to provide feedback for inertial module. The system innovatively proposes the adaptive feature adjustment algorithm, which effectively solves the problem between accuracy and computational cost. The characteristics of LL-SLAM, such as computational efficiency, excellent robustness, absolute trajectory scale, rapid initialization, and high accuracy, can make the system more suitable for UAV flights. We evaluate our system on public benchmarks and UAV flights for pose estimation and time cost compared to other state-of-the-art SLAM systems. In addition, experiments on complex UAV flight tasks show that our system can favorably meet the needs of UAV Autonomous Navigation.