An image-based methodology is presented for the automatic generation of geometric building models at LOD4, incorporating both interior and exterior geometrical information. Existing approaches often focus on simplified geometries for either exteriors or interiors, leading to integration challenges due to data complexity and processing demands. This methodology addresses these challenges by utilizing three structure-from-motion models: one for the building exterior, one for the interior, and one for the entrance. The exterior and interior data are processed separately using planar primitives, and the models are subsequently aligned through a 3D point cloud registration method based on 2D image features. This ensures a unified coordinate system and accurate generation of the LOD4 model. The framework achieved a mean relative error of 3.06% and a mean absolute error of 0.05 m, underscoring its robustness for applications such as numerical modeling, construction management, and structural health monitoring, making it valuable for further advancements in building information models and digital twins.

中文翻译：

---

为建筑物生成 LOD4 模型，以实现 BIM 和数字孪生的自动 3D 建模


提出了一种基于图像的方法，用于在 LOD4 下自动生成几何建筑模型，其中包含内部和外部几何信息。现有方法通常侧重于外部或内部的简化几何图形，由于数据复杂性和处理需求，导致集成挑战。该方法通过使用三个运动结构模型来解决这些挑战：一个用于建筑物外部，一个用于内部，一个用于入口。外部和内部数据使用平面基元分别处理，随后通过基于 2D 图像特征的 3D 点云配准方法对齐模型。这确保了统一的坐标系和准确生成 LOD4 模型。该框架的平均相对误差为 3.06%，平均绝对误差为 0.05 m，凸显了其在数值建模、施工管理和结构健康监测等应用中的稳健性，使其对建筑信息模型和数字孪生的进一步发展具有重要价值。