针对煤矿掘进巷道存在典型非结构化、特征退化、大尺度环境，巷道三维重建易出现位姿估计精度低、累计漂移误差大等问题，提出了一种激光雷达和惯导（Inertial Measurement Unit，IMU）融合的煤矿掘进巷道三维重建方法。该方法通过迭代卡尔曼滤波将激光雷达观测模型的残差函数和IMU预测模型的先验状态偏差紧耦合，经状态更新得到更为精确的后验状态，提升了退化环境下的位姿估计精度。为降低巷道三维模型重建过程中的累计漂移误差，提出基于体素化广义迭代最近点（Voxelized Generalized ICP，VGICP）的回环检测算法，以基于体素的单分布到多分布的方式进行配准，完成对回环帧的选取及精确匹配，实现回环帧的全局位姿校正，有效降低煤矿巷道三维重建的累计漂移误差。相比于A-LOAM、LEGO-LOAM、LINS算法，所提算法在位姿估计精度和全局一致性方面显著提升。公开数据集实验结果表明：所提算法的RPE和APE均方根误差分别为0.2718和0.5008，与其他算法相比分别降低了53.14%、50.97%、48.31%，和50.41%、47.99%、47.49%。开展了3种模拟巷道场景三维重建实验，结果表明所提算法构建的室内长廊模型各区域在长度、宽度和高度方向的误差均在1.2%以内；所提算法构建的煤矿巷道三维模型与真实巷道空间分布一致，总体距离退化误差仅为2.46%，较其他3种算法重建性能分别提升了66.12%、65.30%、70.43%。在煤矿主体实验室掘进巷道进行三维重建实验，结果表明三维重建结果在长度、宽度和高度方向的平均误差百分比分别为0.47%、0.75%和0.67%，可以实现掘进巷道三维精确建模。

In view of the problems of typical unstructured, degraded characteristics and large-scale environment in coal mine tunnel, and the 3D reconstruction of coal mine tunnel is prone to low pose estimation accuracy and large cumulative drift error, a 3D reconstruction method of coal mine tunnel fused with lidar and inertial measurement unit (IMU) is proposed. In this method, the residual function of the lidar observation model and the prior state deviation of the IMU prediction model are tightly coupled through iterative kalman filter, and a more accurate posteriori state is obtained through state update, which provides reliable pose estimation for the degraded environment. In order to reduce the cumulative drift error in the process of tunnel 3D model reconstruction, a loopback detection algorithm based on Voxelized Generalized ICP (VGICP) is proposed. It is registered in a voxel-based single to multi-distribution mode, so as to complete the selection and accurate matching of loopback keyframes, realize the global pose correction of loopback keyframes, and effectively reduce the cumulative drift error of 3D reconstruction of coal mine tunnel. Compared with the A-LOAM and LEGO-LOAM algorithms, the proposed algorithm has significantly improved the accuracy and global consistency of pose estimation. Experimental results on public datasets show that the root mean square errors of RPE and APE of the proposed algorithm are 0.2718 and 0.5008, respectively, which are reduced by 53.14%, 50.97%, 48.31%, and 50.41%, 47.99% 47.49% respectively compared with other algorithms. Two real-world 3D reconstruction experiments are carried out. The results show that the error percentage of each area of the indoor corridor model constructed by the proposed algorithm in the directions of length, width and height was within 1.2%. The three-dimensional model of coal mine tunnel constructed by the proposed algorithm is consistent with the spatial distribution of the real coal mine tunnel. The overall distance degradation error is only 2.46%, which is 66.12%, 65.30% and 70.43% higher than that of the other three algorithms, respectively. Three-dimensional reconstruction experiments are carried out in the roadway excavation in the main laboratory of the coal mine. The results show that the average error percentages in the length, width and height directions are 0.47%, 0.75% and 0.67%, respectively. It can realize the accurate three-dimensional modeling of the tunnel.