基于空间梯度滤波和k近邻聚类融合的巷道点云掌子面分割方法

A Point Clouds Segmentation Method for Tunnel Face Based on Spatial Gradient Filtering and k-means Clustering

针对现有的曲面滤波方法不完全适用于矿山巷道掌子面点云分割的问题,提出了一种融合空间梯度滤波与k近邻欧式聚类的点云分割算法。该算法基于空间梯度滤波降低点云数据中掌子面和巷道壁区域黏连的现象,采用k近邻欧式聚类算法分离不同点云团体并保留掌子面点云分割结果。最后,采用AABB包围盒筛选滤波步骤滤除的点云,并用来补全因过分割而导致的掌子面内部空洞区域。试验选取4组真实矿山巷道雷达点云集合,从巷道形貌、点云密度、姿态角偏差量等不同维度验证所提算法的鲁棒性。试验结果表明,在最优超参数条件下,该算法的精确率、召回率与综合评价指标F1分别达到了97.80%、98.98%和98.38%,均优于所选的对比算法。对比试验证明该算法能够实现复杂曲面形貌条件下的掌子面点云分割,具有精度高、可操作性好、鲁棒性高的特点。

Aiming at the problem that the existing surface filtering methods might partially apply to the point clouds segmentation of the tunnel face of mineral tunnels,a segmentation method for the tunnel face based on spatial gradient filtering and k-means clustering was proposed.The process removed the connection between the tunnel face and the tunnel wall based on a spatial gradient filtering,and the k-means clustering was used to separate different point cloud groups and retain the point clouds segmentation results of the tunnel face.Finally,the AABB bounding box was used to screen the filtered out point clouds to fill in the interior holes of the tunnel face caused by over-segmentation.Four groups of real mine lidar point cloud sets were selected in the test,and the robustness of the proposed algorithm was verified from different dimensions such as tunnel morphology,point cloud density and attitude angle deviation.The test results show that under the optimal hyper-parameter conditions,the accuracy,recall and comprehensive evaluation index F1 of the algorithm reach 97.80%,98.98% and 98.38%,respectively,which are better than the selected comparison algorithm.The comparative experiments show that the algorithm can realize the point clouds segmentation of the tunnel face under the condition of complex surface topography,and has the characteristics of high precision,good maneuverability and high robustness.