Abstract: For temporary roads guided by traffic cones,this paper proposes an improved Delaunay triangulation algorithm to implement road detection in this special scene.YOLOv4 algorithm is used to recognize traffic cones in the images.Besides,the image information and the point cloud information of the traffic cones obtained by laser radar are fused,and the fused traffic cone information is Delaunay triangulated to propose a Delaunay triangulation filtering algorithm and the local optimization strategy,which calculates the weight and the loss value of the Delaunay triangulation in real time according to the change of road conditions and effectively filters out the triangular edges whose sum of the loss values fails to match the conditions.The proposed method reduces the noise constraint in the Delaunay triangulation and effectively implements the rapid planning and real-time updating of lane lines and drivable paths.The real vehicle experiment results show that the average time consumption of the proposed method is 35.4 ms.The absolute trajectory error of the detected path is 0.2 m,and the accuracy is 97%.Compared with the traditional Delaunay triangulation method,the improved method meets the real-time demand,reduces the path detection error,and improves the path detection accuracy.