【摘要】 在大型桥梁工程建设中，实景三维建模技术对智能建造、智能施工与智能管养等领域具有重要应用价值。然而，在复杂且险峻的建设环境下，基于单一数据源的三维实景建模难以真实表达桥梁的三维信息与纹理特征。为解决此问题，提出了一种基于激光雷达点云与视觉图像融合的实景三维建模技术。以渭河管桥主桥桥塔为研究对象，综合倾斜摄影、机载激光扫描、地面激光扫描技术对主桥桥塔进行数据采集，并对所采集的数据进行去噪及坐标系统一化处理；通过全局与局部双阶段配准算法实现异源点云数据的高精度匹配；利用实景建模软件结合倾斜摄影影像实现纹理映射生成桥塔实景三维模型。所建立的实景三维模型完整地表示了结构几何特征，良好地还原了建筑物的真实情况。为了验证建模精度，对比分析了桥塔的三维模型数据与几何实测数据，桥塔实景模型的高程误差集中分布在4.00～4.70 cm,均值为4.32 cm,误差相较结构整体尺寸而言，高程误差/实际高程比值在0.3%以下。这表明所提出方法能较好地实现实景三维建模，能较准确表达桥梁桥塔的复杂几何细节和纹理特征，为大型桥梁建设的三维可视化监测与管理提供了可靠的数据支持。更多还原

【Abstract】 In large-scale bridge engineering, 3D real-scene modeling technology plays a crucial role in intelligent construction, execution, and operation & maintenance. However, in complex and challenging construction environments, 3D modeling that rely on a single data source often struggle to accurately capture the detailed geometry and surface textures of bridge structures, leading to inadequate representations. To address this limitation, an integrated 3D real-scene modeling method was proposed that combines LiDAR point clouds with visual imagery. Taking the main pylon of the Wei River Bridge as a case study, data acquisition was performed by integrating oblique photogrammetry, airborne laser scanning(ALS), and terrestrial laser scanning(TLS). The acquired multi-source data underwent preprocessing, including denoising and coordinate system unification. A two-stage registration algorithm(global-to-local) was then applied to achieve high-precision alignment of the point clouds. Subsequently, texture mapping based on oblique imagery was performed on the registered point cloud data using professional reality modeling software to generate a photorealistic 3D model of the pylon.The applied methodology yielded a real-scene 3D model with precise geometric representation and high-fidelity visual reconstruction of the actual structure.To validate the modeling accuracy, a comparative analysis was conducted between the 3D model data and geometrically measured field data of the pylon. The elevation errors of the real-scene model were predominantly distributed within the range of 4.00 cm to 4.70 cm, with an average error of 4.32 cm. Relative to the overall dimensions of the structure, the ratio of elevation error to actual elevation was less than 0.3%. These results demonstrate that the proposed method can effectively achieve high-fidelity 3D real-scene modeling, accurately capturing the complex geometric details and textural features of bridge pylons. This approach provides reliable data support for 3D visualization, monitoring, and management throughout the lifecycle of large-scale bridge construction projects.更多还原