基于动态边缘化的双目视觉惯性SLAM算法

STEREO VISUAL INERTIAL SLAM ALGORITHM BASED ON DYNAMIC MARGINALIZATION

针对单目视觉惯性SLAM算法鲁棒性不高且尺度恢复困难的问题,提出基于动态边缘化的双目视觉惯性SLAM算法(DM-SVI-SLAM)。前端使用光流法进行特征跟踪,利用预积分计算帧间IMU,后端在滑动窗口内融合单/双目匹配点误差、IMU残差及先验误差构建捆集调整的成本函数,利用动态边缘化策略、Dog-Leg算法提升计算效率,回环检测使用词袋方法对关键帧重定位。通过EuRoC数据集评估系统性能,实验结果表明,对比其他前沿VI-SLAM算法,该算法在精度和鲁棒性方面都具有潜力。

Aiming at the shortcomings of the monocular visual inertial simultaneous localization and mapping(SLAM)algorithm in poor robustness and low accuracy of scale estimation,a stereo visual inertial SLAM algorithm based on dynamic marginalization(DM-SVI-SLAM)is proposed.In the frontend,in order to decrease the cost of computation,we tracked features by the KLT sparse optical flow algorithm and predicted the pose of the current frame by pre-integrating the IMU data from the previous frame to the current frame.Then,in the backend,we constructed bundle adjustment cost function containing monocular and binocular matching point errors,IMU residuals,and prior errors in a tightly coupled sliding window and improved computing efficiency by introducing dynamic marginalization strategy and Dog-Leg algorithm.In addition,in the loop detection,we relocated keyframes by calculating the bag-of-words vector.Finally,the performance of the system was verified with the EuRoC dataset.The experimental results demonstrate that the proposed algorithm is a promising VI-SLAM algorithm compared with other state-of-the-art VI-SLAM algorithms in terms of both accuracy and robustness.