面向煤矿巷道环境的LiDAR与IMU融合定位与建图方法

Coal mine roadway environment-oriented LiDAR and IMU fusion positioning and mapping method

  • 摘要: 针对煤矿井下喷浆表面、对称巷道等引起移动机器人自主导航定位与建图失效问题,提出了一种面向煤矿巷道环境的激光雷达(LiDAR)与惯性测量单元(IMU)融合的实时定位与建图方法。首先对原始点云进行分割,利用IMU预积分位姿去除原始点云非线性运动畸变,并对得到的点云进行线、面特征提取。然后将相邻帧的线、面特征进行匹配,在分层位姿估计过程中融合IMU预积分所得到的位姿初值,减少计算迭代次数,提高特征点匹配的精度,解算出当前帧的位姿。最后向因子图中插入局部地图因子、IMU因子、关键帧因子,对位姿进行优化约束,对关键帧与局部地图进行匹配,通过八叉树结构实现地图构建。为验证所提方法的定位性能与建图效果,搭建了Autolabor 、VLP−16 LiDAR和Ellipse−N IMU的实验平台进行验证,并与LeGO−LOAM、LIO−SAM方法进行定性定量对比分析。结果表明:① 在煤矿巷道环境中,面向煤矿巷道环境的LiDAR与IMU融合的实时定位与建图方法三轴方向的绝对定位误差的均值和中值均小于32 cm;对X轴的位姿估计精度最高,其累计误差为1.65 m,位置偏差为2.97 m,建图效果整体良好,建图轨迹未发生漂移;构建的点云地图在完整性和几何结构真实性方面均有着优秀的表现,可以直观反映巷道环境的实际情况,具有良好的鲁棒性。这是因为点云匹配之后进行了分层位姿估计,多因子优化可有效降低全局累计误差,对轨迹精度和地图的一致性提升具有重要作用。② 在楼道走廊环境中,面向煤矿巷道环境的LiDAR与IMU融合的实时定位与建图方法三轴的误差均小于1.01 m,误差均值为5~15 cm,误差范围小,精度高;累计位置偏差仅为1.67 m;完整性与环境匹配均有良好的性能。这是由于通过增加关键帧因子,插入因子图对其新增节点相关变量进行优化,降低了位姿估计漂移,定位与建图精度相对较高。

     

    Abstract: The failure of autonomous navigation, positioning and mapping of the mobile robot is caused by the shotcrete surface and symmetrical roadway in coal mine. In order to solve this problem, a real-time positioning and mapping method based on LiDAR and IMU fusion is proposed for the roadway environment in the coal mine. Firstly, the original point cloud is segmented. The IMU pre-integration pose is used to remove the nonlinear motion distortion of the original point cloud. The line and surface feature extraction is carried out on the obtained point cloud. Secondly, the line and surface features of adjacent frames are matched. The initial pose value obtained by IMU pre-integration is fused in the hierarchical pose estimation process. The calculation iteration times are reduced, the matching precision of feature points is improved, and the pose of the current frame is solved. Finally, the local map factor, IMU factor and key frame factor are inserted into the factor graph to optimize and constrain the pose. The key frame is matched with the local map, and the map construction is realized through an octree structure. In order to verify the positioning performance and mapping effect of the proposed method, the experimental platforms of Autolabor, VLP-16 LiDAR and Ellipse-N IMU are built. The qualitative and quantitative comparison between the proposed method and LeGO-LOAM and LIO-SAM methods is carried out. The results show the following points. ① In the coal mine roadway environment, the average and median of the absolute positioning error in the three axes direction of the real-time positioning and mapping method based on LiDAR and IMU fusion are less than 32 cm. The position and attitude estimation precision in the X-axis is the highest, with a cumulative error of 1.65 m and a position deviation of 2.97 m. The overall mapping effect is good, and the mapping track does not drift. The point cloud map constructed has excellent performance in integrity and geometric structure authenticity. The map can directly reflect the actual situation of the roadway environment, and has good robustness. This is because hierarchical pose estimation is performed after point cloud matching. The multi-factor optimization can effectively reduce the global cumulative error, which plays an important role in improving track precision and map consistency. ② In the corridor environment, the three-axis error of the real-time positioning and mapping method based on LiDAR and IMU fusion for the coal mine roadway environment is less than 1.01 m. The average error is 5~15 cm, with small error range and high precision. The accumulated position deviation is only 1.67 m. Integrity and environment matching have good performance. This is because by adding keyframe factors and inserting factor graphs to optimize the related variables of the newly added nodes, the drift of pose estimation is reduced. The positioning and mapping precision is relatively high.

     

/

返回文章
返回