Urban digital twins are becoming essential for transportation applications, demanding precise geometric, semantic, and topological data. However, existing transportation infrastructure information is typically available in 2D formats, while many applications require accurate 3D representations. Existing 3D representations, such as point cloud data, often lack integrated semantic information. This paper addresses this gap by presenting a novel method for the automatic transformation of semantic 2D lane models into 3D CityGML representations. The transformation process comprises three main phases: (1) Point cloud data processing: Noise and irrelevant structures are removed, retaining essential 3D lane features, and elevation information is derived by converting the point cloud data into digital elevation models (DEMs); (2) Segmentation and smoothing: Extracted DEMs undergo segmentation, noise removal, and refinement to ensure geometric continuity; and (3) Transformation and postprocessing: The semantic 2D lane models are integrated with the processed DEMs through elevation interpolation, followed by refinement and transformation into 3D CityGML representations. Compared to existing methods, the proposed method delivers more realistic and comprehensive 3D lane models while maintaining efficiency. A case study in Munich, Germany, demonstrates the algorithm’s effectiveness in addressing challenges in complex scenarios including tunnels and bridges. The paper concludes by discussing encountered challenges and proposing future research directions to advance the integration of 2D and 3D transportation infrastructure information.