In order to accurately extract the road edge in front of the intelligent vehicle in a variety of road environments, an approach of road edge detection based on 3D LIDAR is proposed. The algorithm firstly uses the random sampling consistency algorithm to quickly segment the road area, filter out most of the non-ground data, and improve the processing speed of the subsequent steps. An extraction algorithm based on the unrelated graph neighborhood relationship is proposed, which is characterized by multi-feature, wide-threshold and multi-level curb feature, to improve the accuracy of the roadside detection by setting a variety of roadside geometric features and wider thresholds. Then the candidate points of road edge are obtained by bidirectional scanning line search algorithm. Clustering analysis and denoising are carried out according to the characteristics of density and global continuity of road edge. Finally, the road edge is fitted by quadratic curve.The results show that the algorithm can effectively identify the structural straight and curved road edges under the occlusion of vehicles, pedestrians and obstacles. The accuracy of the algorithm is higher than 86%, and the detection road width error is less than 0.19 m, which verifies the robustness and accuracy of the algorithm.