The efficient detection and extraction of lane lines is one of the key technologies that urgently need to be overcome in the field of autonomous driving. Many detection algorithms based on visual solutions have certain limitations due to the characteristics of image data, such as the impact of weather lighting on imaging quality and the difficulty of considering both curved and straight roads. This article proposes an algorithm for automatic extraction of lane lines by combining the advantages of 3D laser point cloud and road knowledge rules. Firstly, road surface points are obtained by enhancing the elevation differences of road boundaries multiple times. Secondly, the Isodata algorithm is simplified to adaptively obtain the threshold for intensity filtering. Then, the random sample consensus algorithm is used to detect straight line clusters and obtain candidate lanes. The candidate lanes are mapped into 2D vectors and correct lane lines are extracted through inter-class distance constraints. Finally, the vector topology consistency based on adjacent key feature point pairs is used to reconstruct the lane topology and obtain complete and meaningful lane lines in the real world. The algorithm achieves 92.46% recall, 94.79% accuracy, and 92.41% overall evaluation index with up to 5~6 lane lines. Experimental results prove effectiveness and feasibility of the method.