Typical curb recognition algorithms have difficulty in balancing real-time performance and reliability. In this paper, with a multi-line LiDAR used, a curb recognition algorithm based on geometric features and 3D point cloud features of curb areas is proposed, which reaches a tradeoff between real-time performance and reliability. Faced with the large amount of point cloud data, the algorithm firstly proposes a ground segmentation method based on RANSAC algorithm, filtering out the ground points in the preset region of interest, and then the orderly rasterization of the remaining disordered points is carried out for matching and screening curb areas according to the curb’s geometric characters and the points’ distribution feature. After which, the least square method fused with RANSAC is proposed to achieve the robust fitting of curb curve. Experiments show that the recognition accuracy of the algorithm is more than 95% in both straight and bend scenes, and the time-consuming is less than 15ms, which indicates the good accuracy and real-time performance of the proposed algorithm. The algorithm can effectively identify road curb, thus providing a theoretical reference and method basis for intelligent vehicle driving area recognition and its’ control.