Trees play a crucial role in urban environment, offering distinct ecological and aesthetic values. Fine-grained urban tree models hold valuable potential for urban landscape planning and green space management. Consequently, in recent years, how to reconstruct detailed tree models using digital twin technology has become a focal point of interest. Point cloud data has become a major source for tree modeling because of its unique capability to represent objects’ geometry. However, current methods for tree reconstruction primarily concentrate on individual trees, and necessitate high-quality point cloud data that is arduous to acquire on a large scale within complex urban settings. Furthermore, adequately storing and managing tree models of a large scale is another vital challenge to address. In response to these challenges, we propose a novel approach for 3D modeling and parameterization of large-scale urban trees based on point clouds acquired from consumer-grade mobile laser scanning (MLS) and UAV laser scanning (ULS) platforms. Our pipeline encompasses several key techniques: tree extraction, adaptive tree modeling, and model parameterization. To validate our approach, we collected MLS and ULS data covering an area of 36,400 $m^2$. Using the proposed pipeline, we achieved large-scale tree modeling and lightweight model representation with a success rate of 96%. Both qualitative and quantitative evaluations have proved the effectiveness of our reconstruction method in terms of visual quality and estimation of tree structure parameters. The generated tree models, amenable to lightweight representation, facilitate integration and contribute to the advancement of digital urban forest construction, management, and applications.

树木在城市环境中发挥着至关重要的作用，具有独特的生态和美学价值。细粒度的城市树木模型对于城市景观规划和绿地管理具有宝贵的潜力。因此，近年来，如何利用数字孪生技术重建详细的树木模型成为人们关注的焦点。点云数据因其独特的表示对象几何形状的能力而成为树建模的主要来源。然而，当前的树木重建方法主要集中于单棵树木，并且需要高质量的点云数据，而这些数据很难在复杂的城市环境中大规模获取。此外，充分存储和管理大规模的树模型是另一个需要解决的重要挑战。为了应对这些挑战，我们提出了一种基于从消费级移动激光扫描 (MLS) 和无人机激光扫描 (ULS) 平台获取的点云对大型城市树木进行 3D 建模和参数化的新方法。我们的流程包含多项关键技术：树提取、自适应树建模和模型参数化。为了验证我们的方法，我们收集了覆盖 36,400 个区域的 MLS 和 ULS 数据${米}^2$。使用所提出的管道，我们实现了大规模树建模和轻量级模型表示，成功率为 96%。定性和定量评估都证明了我们的重建方法在视觉质量和树结构参数估计方面的有效性。生成的树模型适合轻量级表示，有利于集成，有助于推进数字城市森林建设、管理和应用。