ISSN 1000-3665 CN 11-2202/P
    徐志华,郭戈,孙钱程,等. 改进的区域生长算法在三维激光点云识别岩体结构面中的应用[J]. 水文地质工程地质,2024,51(0): 1-13. DOI: 10.16030/j.cnki.issn.1000-3665.202303051
    引用本文: 徐志华,郭戈,孙钱程,等. 改进的区域生长算法在三维激光点云识别岩体结构面中的应用[J]. 水文地质工程地质,2024,51(0): 1-13. DOI: 10.16030/j.cnki.issn.1000-3665.202303051
    XU Zhihua, GUO Ge, SUN Qiancheng, et al. An improved region growing algorithm in 3D laser point cloud identification of rock mass structural plane[J]. Hydrogeology & Engineering Geology, 2024, 51(0): 1-13. DOI: 10.16030/j.cnki.issn.1000-3665.202303051
    Citation: XU Zhihua, GUO Ge, SUN Qiancheng, et al. An improved region growing algorithm in 3D laser point cloud identification of rock mass structural plane[J]. Hydrogeology & Engineering Geology, 2024, 51(0): 1-13. DOI: 10.16030/j.cnki.issn.1000-3665.202303051

    改进的区域生长算法在三维激光点云识别岩体结构面中的应用

    An improved region growing algorithm in 3D laser point cloud identification of rock mass structural plane

    • 摘要: 交错分布的结构面构成了岩体中的薄弱部位,准确高效的岩体结构面识别和特征信息提取可为岩体稳定性评价提供重要依据。三维激光扫描技术可以极大地提高结构面勘测效率和精度,但目前主流的点云分析算法存在结构面边缘识别模糊、点云分割准确性不能满足结构面特征信息提取精度的问题。因此,考虑岩体结构面点云位置与其邻域的空间关系,利用KD-tree数据结构进行最邻近搜索的体素下采样,在稳健随机Hough变换的基础上改进了区域生长算法,通过多特征值对区域生长分割参数进行修正,依据点云法向量差值和特征终值进行结构面分割,实现了结构面产状、间距、延展度信息的提取。研究结果表明:与传统的主成分分析法和随机抽样一致法相比,在室内块体模型组成的24个结构面中,该方法在相同区域具有更高的识别率和准确率,既能在复杂变化的平面区域保证数据的完整识别,也能在平面的尖锐位置较好地分割边缘点云。利用该方法可以将24个结构面分为6组,并在识别数据中获取对应的结构面特征信息,与实际测量结果相比,角度信息误差约为1°,距离信息误差1 cm以内。该方法在长江干流蟒蛇寨斜坡岩体中成功识别出3组结构面同时计算各组结构面间距与延展度信息,并采用赤平投影图分析不同结构面组对斜坡稳定性的影响。所提出的方法在室内模型及现场斜坡验证效果良好,可以为岩体结构面识别分割提供稳定且有效的技术支撑。

       

      Abstract: The rock mass structural plane constitutes the weakest part of the rock mass. Accurate and efficient identification of rock mass structural plane and extraction of characteristic information can provide an important basis for the rock mass stability evaluation. 3D laser scanning technology can greatly improve the efficiency and accuracy of structural surface survey; however, the current mainstream point cloud analysis algorithms exist the problems that the edge recognition of structural surfaces is blurred and the accuracy of point cloud segmentation cannot meet the accuracy of structural surface feature information extraction. Considering the spatial relationship between the position of the point cloud of the rock mass structural plane and its neighborhood, the region growth segmentation parameters were corrected by multiple eigenvalues. The KD-tree data structure was used to perform the nearest neighbor search. The voxel was sampled, and the structural plane was segmented to realize the extraction of the structure plane occurrence, spacing, and extension information, based on the normal vector difference of the point cloud and the characteristic final value. The effectiveness of this method in structural plane identification was also verified by indoor models. The results show that compared with the traditional Principal Component Analysis method and Random Sample Consensus method, this method has a higher recognition rate and accuracy in the same area among the 24 structural planes composed of indoor block models. It can not only ensure the complete recognition of data in the complex and changing plane area, but also better segment the edge points in the sharp position of the plane. Using this method, 24 structural planes can be divided into 6 groups, and the corresponding structural plane feature information can be obtained. Compared with the actual measurement results, the angle information error is approximately 1°, and the distance information error is within 1 cm. This method identified three groups of structural planes in the Mangshezhai slope rock mass successfully in the main stream of the Yangtze River. The method proposed in this study has a good verification effect on indoor model and field slope, which can provide robust and effective technical support for the identification and segmentation of rock mass structural plane.

       

    /

    返回文章
    返回