Abstract: The demands for urban road feature extraction and 3D reconstruction are increasingly prominent in modern urban traffic management, autonomous driving, urban planning, and geographic information systems. However, existing methods are hindered by low processing efficiency, high human intervention, and excessive optimization for specific environments, which makes those methods insufficient to meet current requirements. This paper proposed a method to automatically extract road features from lidar point cloud data and establish a 3D model, effectively handling large-scale complex environmental data. First, road boundary lines were determined by extracting curbstone descriptors. Afterwards, a geographic reference image of the point cloud data was generated, extracting refined road markings. Following that, rod-shaped objects were detected under smoothness constraints, and street lights and roadside trees were distinguished using a classification algorithm. Finally, parameters required for the 3D reconstruction of the road model were analyzed, and a continuous quadrilateral reconstruction method was proposed, achieving 3D reconstruction of road elements. Experimental results show that the evaluation accuracy of proposed method has reached 92% in road point cloud data target extraction, which verifies its effectiveness.