基于三角词袋回环检测的激光惯性SLAM算法

LiDAR-inertial SLAM algorithm based on triangle bag of words loop closure detection

回环检测是减少激光惯性同步定位与建图(SLAM)位姿漂移的有效方法,而回环检测的精度和速度是其能否被应用于SLAM的关键因素。基于此,提出了一种基于三角词袋回环检测的激光惯性SLAM算法。首先,通过激光点云的LinK3D特征生成三角描述符,使用三角描述符构建三角词袋,实现实时位置识别与六自由度回环位姿估计。其次,将LinK3D特征用于帧到帧的点云配准,与惯性测量装置(IMU)预积分相结合,实现精确鲁棒的帧间位姿估计。在KITTI数据集上的实验结果表明,与LIO-SAM算法相比,所提SLAM算法的帧间位姿估计方法更加鲁棒,轨迹的平均均方根误差减少29.79%,每次回环约束的平均耗时减少93.53%。实测实验结果表明,与LIO-SAM算法相比,所提算法每次回环约束的平均耗时减少85.15%,室外长距离实验的绝对轨迹误差的均方根误差减少84.36%。

Loop closure detection is an effective strategy to reduce the pose drift of lidar-inertial simultaneous localization and mapping(SLAM),and the accuracy and speed of loop closure detection are key factors for its application in SLAM.Based on this,a lidar-inertial SLAM algorithm based on triangle bag of words loop closure detection is proposed.Firstly,a triangle descriptors are generated by the LinK3D features of the laser point clouds,and a triangle word bag is constructed by using the triangle descriptors to achieve real-time position recognition and six-degree-of-freedom loop pose estimation.Secondly,LinK3D features can also be used for frame to frame point cloud registration,combined with inertial measurement unit(IMU)pre-integration to achieve accurate and robust interframe pose estimation.The experimental results on the KITTI dataset show that compared with the current advanced LIO-SAM algorithm,the proposed SLAM algorithm has a more robust interframe pose estimation,the average root mean square error of the output trajectory is reduced by 29.79%,and the average time required for each loop constraint is reduced by 93.53%.The field experimental results show that compared with LIO-SAM,the proposed algorithm reduces the average time required for each loop constraint by 85.15%,and the root mean square error of the absolute trajectory error in outdoor long-distance experiments is reduced by 84.36%.