大直径钻孔救援是一种有效的矿山灾害应急救援方法，通过向人员被困位置施工大直径救援井，采用提升舱将被困井下的人员通过大直径救援井安全可靠提升至地面。大直径救援井的结构参数及其变化对提升舱的通过性能具有重要影响，实时获取大直径救援井的结构参数是大直径钻孔救援科学操作与决策的基础。为了满足大直径救援井结构参数的监测需求，研制了基于双目结构光原理的Intel RealSense D435i深度相机和ROS操作系统的大直径救援井结构参数检测与三维模型重建系统。针对大直径救援井井筒纹理光滑不明显的问题，对比分析了SIFT、SURF、ORB三种图像特征点提取算法提取大直径救援井井筒图像特征点的效果；采用ICP-PNP混合算法，估计深度相机的位姿，并根据相邻帧之间的运动误差筛选关键帧，优化三维点云模型；最后基于大直径救援井提升过程模拟试验平台进行了大直径救援井三维模型的重建试验。研究结果表明：在大直径救援井井筒纹理结构不明显情况下，相同时间内SURF算法具有更好的特征点提取效果；根据相邻帧之间的运动误差筛选关键帧图像，可以稳定地构建大直径救援井三维模型；对大直径救援井三维点云模型进行截面采样，与实际井筒半径误差在2%以内；基于最小二乘圆法对大直径救援井三维点云模型的采样截面进行圆拟合，与实际井筒半径误差在0.4%以内。重建的大直径救援井三维模型能够准确地反映实际大直径救援井的结构参数，为地面应急救援提升的科学操作和决策提供了保障。

Large-diameter borehole rescue is an effective emergency rescue method for mine disasters. Through the large-diameter rescue well that is constructed to the place where the trapped personnel is located, the trapped personnel can be lifted safely and reliably to the ground by a hoisting capsule. The structural parameters of a large-diameter rescue well and their changes have significant impact on the passing performance of the hoisting capsule. The real-time acquisition of the structural parameters of large-diameter rescue well is the basis of scientific operation and decision-making of large-diameter borehole rescue. In order to satisfy the requirements for monitoring of the structural parameters of large-diameter rescue well, the structural parameter detection and 3D model reconstruction system of large-diameter rescue well was developed based on the Interl RealSense D435i depth camera in the principle of binocular structured light and ROS operation system. To solve the problem that the texture of large-diameter rescue well is smooth and not noticeable, the SIFT, SURF and ORB algorithms were compared and analyzed in terms of the borehole image feature point extraction effect of large-diameter rescue well. Besides, the pose of depth camera was estimated with the ICP-PNP algorithm, and the key frames were screened according to the motion errors between adjacent frames, so as to optimize the 3D point cloud model. Finally, the 3D remodeling test of the large-diameter rescue well was carried out based on the simulation test platform of the lifting process in the large-diameter rescue well. The research results show that: The SURF algorithm has better effect on feature point extraction in the same time when the wellbore texture is not noticeable. The 3D model of large-diameter rescue well can be constructed stably by screening the key frame images according to the motion errors between adjacent frames. The cross-section sampling was performed on the 3D point cloud model of the large-diameter rescue well, showing that the error is within 2% from the actual wellbore radius. In addition, the sampled section of the 3D point cloud model of the large-diameter rescue well was fitted to the circle with the least square circle method, which indicates the error between the fitted circle and the actual wellbore radius is less than 0.4%. The reconstructed 3D model of large-diameter rescue well could accurately reflect the structural parameters of the actual large-diameter rescue well and guarantee the scientific operation and decision-making of surface emergency rescue lifting.