面向城市更新的点云-BIM几何语义一体化既有建筑三维重建方法

Point Cloud-Driven Building Information Modeling Reconstruction Method with Geometric-Semantic Integration for Existing Buildings in Urban Regeneration

摘要:

为解决基于点云的建筑信息模型（BIM）重建耗时费力且质量稳定性不足的问题，提出一种针对既有建筑主体室外部分的点云?BIM几何语义一体化三维重建方法。利用Mask3D算法对建筑主体室外点云进行语义实例分割与细粒度分类；结合二维?三维视觉融合与拓扑约束优化机制，提出一套室外构件图元参数解耦算法以准确提取参数；通过异构数据格式自适应转换接口实现图元参数结构化处理，设计三阶段建模方法，完成室外BIM模型自动化快速重建。案例验证结果表明，该方法能高效重建出集精准的位置、尺寸等几何信息和丰富属性、拓扑关系等语义信息于一体的BIM模型。两组案例中对6类主要室外构件的重建率达96.7%和93.6%，外墙重建精度达厘米级，构件间拓扑关系重建率在95%以上。

Abstract:

This paper proposes a novel building information modeling （BIM） reconstruction methodology with geometric-semantic integration for exterior components of existing buildings， addressing the critical challenges of labor-intensive workflows and inconsistent quality in point cloud-based BIM reconstruction with urban renewal contexts. The Mask3D algorithm is utilized to achieve semantic instance segmentation and fine-grained classification of exterior point clouds. A series of component-oriented parametric decoupling algorithms enable precise parameter extraction through 2D-3D vision fusion integrated with topological constraint optimization. An adaptive heterogeneous data schema mapping interface performs structured processing of primitive parameters. A three-phase modeling methodology is designed to accomplish automated， rapid reconstruction of exterior BIM models. Case validation results indicate that this method achieves efficient reconstruction of BIM models integrating both precise geometric information （e.g.， position and dimensions） and rich semantic information （e.g.， material attributes and topological relationships）. Quantitative analyses demonstrate reconstruction success rates of 96.7% and 93.6% across six exterior component categories， with wall reconstruction precision reaching centimeter-level accuracy and inter-component topological relationship retention rates exceeding 95%.