针对目前泥水盾构筛分岩渣体积监测分析难度高、准确度难以保证和依赖现场人员经验等问题，提出一种基于双目视觉点云的泥水盾构筛分岩渣体积智能监测技术。依托青岛胶州湾第二海底隧道盾构区间，设置多组试验工况，使用双目景深相机获取堆积岩渣目标的点云数据，并对点云模型进行处理计算： 基于RANSAC算法进行点云平面检测与剔除，初步获得较为精准的点云数据，通过姿态变换使点云数据匹配选定的地面坐标系； 基于ICP算法进行点云配准，以此获得横向堆积多土堆模型； 基于高斯滤波的统计学方法进行离散点降噪； 基于Delaunay算法将点云数据三角剖分后进行体积计算，验证分析算法的准确性。结果表明，筛分岩渣体积监测算法在合理的参数配置下能够达到较高的准确率，试验结果平均准确率高于95%，每50环监测平均相对误差稳定在3.94%~5.38%，最小相对误差仅0.86%。
The volume of screened rock slag from slurry shield tunneling is a critical parameter reflecting the overbreak and underbreak conditions of slurry shield tunneling face. It is hard to monitor and analyze with unstable measurement accuracy, and heavily relies on personnel’s experience. To address these challenges, an intelligent monitoring method based on binocular-vision point cloud is proposed for intelligent monitoring of volume of screened rock slag from slurry shield tunneling. Based on a case study of shield tunneling section of the Qingdao Second Jiaozhou Bay Subsea Tunnel, multiple groups of field test conditions are designed, a binocular depth camera is used to collect point cloud data of the accumulated rock slag target, and the point cloud model is systematically processed and calculated. These processes include: (1) The random sample consensus algorithm is applied to detect and eliminate ground plane point clouds, obtaining preliminarily refined target point cloud data. (2) Pose transformation is performed to match the point cloud data with the selected ground reference coordinate system. (3) The iterative closest point algorithm is adopted for point cloud registration to construct a complete model of multiple horizontally accumulated rock slag. (4) The Gaussian filtering combined with a statistical outlier removal method is used to denoise discrete noise points. (5) The volume of the rock slag accumulation is calculated through Delaunay triangulation of the processed point cloud, and the accuracy of the proposed algorithm is verified and analyzed. The test results show that the proposed monitoring algorithm for calculating screened rock slag volume achieves high accuracy under rational parameter configurations. The average accuracy of the test results is higher than 95%, the average relative monitoring error for every 50 rings is stable in the range of 3.94% to 5.38%, and the minimum relative error is only 0.86%.
