The Improved Optimization Algorithm for UAV SLAM in Visual Odometry-Pose Estimation

Pose estimation has always been a key research field in SLAM. The absolute positioning of unmanned aerial vehicle (UAV) in large and complex scenarios is easily affected by environmental factors. When there are external image occlusions and interference with moving objects, the impact on the visual-IMU-based UAV is significant, resulting in poor positioning accuracy and even failure, and IMU is prone to drift. Therefore, this paper proposes a framework based on visual-IMU fusion and distributed extend Kalman filter (DEKF). First, when there are overlapping scene detection and relocation tasks, the visual-IMU SLAM system is applied to the visual odometry (VIO) system through a separate mapping thread, and the IMU information is used to estimate the attitude change of the UAV in a short time when the visual information in the agent is lost. Due to the problem of IMU cumulative drift in the system, the traditional EKF algorithm may lead to the accumulation of a priori error, thus affecting the global optimization results. Therefore, the DEKF algorithm is designed to reduce the IMU drift when visual information is missing. Finally, a case study is conducted using the MH01 open data set to prove the effectiveness of the proposed algorithm.