基于机器学习的城镇满水排水管网三维模型重建技术  

3D Reconstruction Technique for Urban Full Flow Sewer Networks Using Machine Learning

作　　者：封小宇 朱洁明 胡臻臻 张人航 FENG Xiaoyu;ZHU Jieming;HU Zhenzhen;ZHANG Renhang(School of Information and Electrical Engineering,Hangzhou City University,Hangzhou 310027,China;AutoSubsea Vehicles Inc,Shanghai 201306,China;Smart Rovering System Co.,Ltd.,Shanghai 201306,China)

机构地区：[1]浙大城市学院信息与电气工程学院,杭州310027 [2]上海遨拓深水装备技术有限公司,上海201306 [3]上海智曳自动系统有限公司,上海201306

出　　处：《智能物联技术》2025年第1期120-124,共5页Technology of Io T& AI

基　　金：上海市促进产业高质量发展专项资金“两业融合发展”方向项目(SY-LYRH-01-23-0100)。

摘　　要：针对城镇满水排水管网的三维模型重建问题,提出一种基于声呐点云数据的高精度建模方法。首先,预处理原始点云数据,包括去噪、补全和优化,并引入PointMixup技术实现数据增强,以提升模型训练的健壮性和泛化能力;其次,采用基于梯度分析的权重分配法与随机森林模型结合对满水排水管道进行形状拟合,以准确还原管道几何特征;最后,通过提取多分辨率信息,结合Poisson重建技术,在拟合管道形状和原始点云数据的双重约束下,生成细节丰富的三维排水管网模型。以实测水下机器人搭载断面声呐获取的点云数据为实验对象,验证了所提方法的有效性。实验结果表明,该方法能够完整提取管道内部沉积物和水面特征信息,精准重建管道三维结构,同时清晰展示管道的功能性缺陷,具有较高的工程应用价值。To address the challenge of three-dimensional reconstruction for fully submerged urban drainage pipeline networks,this paper proposes a high-precision modeling method based on sonar point cloud data.First,the raw point cloud data is preprocessed through denoising,completion,and optimization.PointMixup technology is introduced to perform data augmentation,enhancing the robustness and generalization capability of model training.Subsequently,a gradient-based weight allocation method is combined with a random forest model to fit the shape of fully submerged drainage pipelines,ensuring the accurate restoration of their geometric characteristics.Finally,multi-resolution information is extracted and integrated with Poisson reconstruction techniques to generate a highly detailed 3D drainage pipeline network model under dual constraints of the fitted pipeline shape and the original point cloud data.The proposed method was tested using point cloud data acquired by underwater robots equipped with cross-sectional sonar in real-world experiments.The results demonstrate its effectiveness.The method successfully captures detailed information on internal sediment and water surface features within the pipelines,accurately reconstructing their 3D structures.Additionally,it provides a clear depiction of functional defects in the pipelines,highlighting its significant engineering application value.

关 键 词：声呐点云数据 机器学习 城镇排水管网 点云分类 规则化处理 三维重建 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]