AbstractRoad curb is one of the important components of road information, and its high‐precision information is significant for the development of autonomous driving, intelligent transportation and smart cities. A mobile laser scanning (MLS) system can acquire high‐precision and high‐density road three‐dimensional (3D) point clouds data, which has the advantages of high efficiency, low cost and non‐contact. However, how to extract accurate road information from the massive and disordered point clouds is one of the current research priorities and difficulties. This paper presents a new method to extract the road curbs from the MLS point clouds. The proposed method mainly includes three steps: pre‐processing, road curbs extraction and vectorisation. Pre‐processing obtains the ground, including road subsection and ground identification. Road curbs are first quantitatively represented by the rotation‐invariant version of the local binary pattern (LBPROT) values in three dimensions, including spatial elevation mode, spatial dispersion mode and spatial shape mode, and then they are extracted by a multidimensional LBPROT features semantic recognition model. Vectorised road curb polylines are connected by accurate road curbs points, which are obtained through simplification and denoising. The proposed method was tested on two large‐scale datasets collected from arterial roads and expressways, respectively. The precision of the results was > 95%, recall was > 90% and the F1 score was > 0.93. The experimental results show that the proposed method can effectively extract road curbs in different environments and has robust adaptability.