摘要
建筑物三维重建是遥感领域的一项基本任务。由于机载LiDAR点云稀疏、不完整、有噪点,传统重建方法难以从中重建出完整水密的模型。同时,建筑物模型的表面紧凑程度也限制了其应用的广泛性。针对该问题,提出了一个从LiDAR点云重建紧凑建筑物模型的网络框架。该架构包含两个模块:一是基于隐式表示的表面提取模块,该模块通过隐式神经网络从点云在提取到建筑物的基本结构信息并生成表面网格模型。一是基于结构感知的网格优化模块,该模块对模型各平面的拓扑进行改善并重建出轻量级建筑物模型。利用苏黎世建筑物数据集对所提出的框架进行了验证,与现有方法相比,该框架重建结果具有更高的精度。
Three-dimensional(3D)reconstruction of buildings is a basic task in remote sensing.Since airborne LiDAR point clouds are sparse,incomplete,and noisy,it is difficult to reconstruct a complete and watertight model from them through traditional reconstruction methods.Meanwhile,the surface compactness of the building model also limits its application.To solve these problems,we propose a network framework for reconstructing compact building models from LiDAR point clouds.The architecture contains two modules:One is the surface extraction module based on implicit representation,which extracts the basic structural information of the building through the implicit neural network and generates the surface mesh model accordingly.The other is a grid optimization module based on structure awareness,which improves the topology of each plane and finally reconstructs a lightweight model.The Zurich building dataset is used to verify the proposed framework.The reconstruction results have higher accuracy than the existing methods.
作者
汪子璐
闫奕名
WANG Zilu;YAN Yiming(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)
出处
《佳木斯大学学报（自然科学版）》
CAS
2023年第5期99-103,共5页
Journal of Jiamusi University：Natural Science Edition
基金
国家自然基金面上项目(62071136)。
关键词
建筑物
三维重建
隐式表示
网格简化
building
3D reconstruction
implicit representation
mesh simplification
