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

作　　者：徐晓苏[1,2] 何宇明 XU Xiaosu;HE Yuming(Key Laboratory of Micro-inertial Instrument and Advanced Navigation Technology,Ministry ofEducation,Nanjing 210096,China;School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]微惯性仪表与先进导航技术教育部重点实验室,南京210096 [2]东南大学仪器科学与工程学院,南京210096

出　　处：《中国惯性技术学报》2024年第9期898-906,917,共10页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金项目(61921004)。

摘　　要：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%.

关 键 词：同步定位与建图 回环检测 词袋模型 点云配准 

分 类 号：TP212.9[自动化与计算机技术—检测技术与自动化装置]