基于李群的无迹卡尔曼滤波视觉SLAM算法

An Unscented Kalman Filter Visual SLAM Algorithm Based on Lie Group

视觉SLAM算法的理论框架已经十分完备,但是在实际应用中导航准确性还有待改善。基于此问题,提出了基于李群的无迹卡尔曼滤波(UKF⁃LG)视觉SLAM算法,优化了传统无迹卡尔曼滤波(UKF)的系统状态,把UKF的系统状态用李群表示,并构建视觉惯性紧耦合模型。在Euroc数据集下对包含该算法在内的5种滤波法进行了仿真对比,其中,L⁃UKF⁃LG算法比传统的UKF算法有更低的位置和姿态的均方根误差(RMSE)值,有效改善了导航定位准确性。

The theoretical framework of visual SLAM algorithm has been very complete,but the navigation accuracy needs to be improved in practical applications.This paper proposes a visual SLAM algorithm for unscented Kalman filter(UKF⁃LG)based on Lie group.The algorithm optimizes the system state of traditional unscented Kalman filter(UKF).And the system state of unscented Kalman filter is represented by Lie group.And a visual inertial tight coupling model is built and compared with five filtering methods including the algorithm under the Euroc data set.Among them,the L⁃UKF⁃LG algorithm has a lower position and posture root mean square error(RMSE)value than the traditional UKF algorithm,effectively improves the accuracy of navigation and positioning.