基于体素层次聚类和纵向约束的变电站电气设备点云数据降噪方法  

Denoising method for point cloud data of substation electrical equipment based on voxel hierarchical clustering and longitudinal constraints

作　　者：郭建龙 李军锋 冯伟夏 薛江 熊山 苏锐师 GUO Jianong;LI Junfeng;FENG Weixia;XUE Jiang;XIONG Shan;SU Ruishi(Training and Evaluation Center of Guangdong Power Grid Co.,Ltd.,Guangzhou 510520,China)

机构地区：[1]广东电网有限责任公司培训与评价中心,广东广州510520

出　　处：《武汉大学学报(工学版)》2025年第1期111-120,共10页Engineering Journal of Wuhan University

基　　金：广东电网有限责任公司科技项目资助(编号:038700KK52170007)。

摘　　要：针对目前点云降噪方法降噪效果差、稳定性不高及泛化能力弱导致设备有效点云提取的准确性低且影响三维模型精度的问题,提出了基于体素层次聚类和纵向约束的降噪方法。首先对变电站三维点云场景按不同高度进行分层区间划分,并将不同高度区间的变电站三维点云场景降维到二维平面,采用基于体素划分的层次聚类方法对点云进行快速聚类;然后,结合电气设备的结构特点,利用设备点云数据与噪声点云数据在纵向连续性上的差异,提出了纵向约束降噪方法,即通过建立纵向约束关系去除点云数据的残余噪声。将所提方法在某220 kV变电站变电设备三维模型建模中应用,并对其降噪效果进行分析。应用与分析结果表明,所提方法具有点云噪声识别误差率低、设备点云几何信息完整性好以及不同点云密度处理的稳定性高等特点。所提方法具有较强的泛化能力,适用于不同电气设备点云数据的提取,能有效提高三维模型的建模精度和效率。In view of the problems of poor noise reduction effect,low stability and weak generalization ability of current point cloud noise reduction methods,which lead to the low accuracy of equipment effective point cloud extraction and affect the accuracy of 3D models,a noise reduction method based on voxel hierarchical clustering and longitudinal constraints is proposed in this paper.Firstly,the three-dimensional point cloud scene of the substation is divided into layers and intervals according to different heights,and the dimensionality of the three-dimensional point cloud scene of substation in different height intervals is reduced to a two-dimensional plane.The hierarchical clustering method based on voxel division is used to cluster the point clouds quickly.Then,based on the structural characteristics of electrical equipment,a longitudinal constraint denoising method is proposed by utilizing the difference in longitudinal continuity between equipment point cloud data and noise point cloud data,so that the residual noise of the point cloud data is removed by establishing a longitudinal constraint relationship.The method proposed in this paper has been applied in the 3D modeling of the substation equipment of a 220 kV substation,and the noise reduction effect is analyzed.The application and analysis results indicate that the proposed method has the characteristics of low error rate of point cloud noise recognition,good integrity of equipment point cloud geometric information,and high stability of cloud density processing at different points.The proposed method has strong generalization ability and is suitable for extracting point cloud data from different electrical equipment,which can effectively improve the accuracy and efficiency of modeling.

关 键 词：点云降噪 层次聚类 纵向约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]