Abstract: In order to solve the problems of difficulty in setting multiple feature thresholds and low generality in road extraction based on airborne light detection and ranging (LiDAR) point cloud, a random forest classification model was used to extract road point cloud and then obtain road center line. Firstly, the ground point cloud was obtained by progressive cryptography triangulation filtering. According to the characteristics of mountain roads, the slope, roughness, height difference variance, point density and reflection intensity of each point in the neighborhood of the ground point cloud were calculated, and the classification characteristics of the component points were calculated. Then, positive and negative samples were collected manually to train the random forest classification model of point cloud. The ground point cloud was classified by the model to get the initial road point cloud. And then, the road point cloud was rifined through the algorithm of density-based spatial clustering of application with noise (DBSCAN). Finally, the road point cloud was vectored to obtain the road center line. The results show that the accuracy of road point cloud extraction is 95.29%, the integrity rate is 92.96%, and the extraction quality is 88.88%, respectively. This method can solve the problem of difficult to determine multiple thresholds, and can extract the mountain road point cloud with high precision, and then obtain the effective road center line, which has certain reference value for the study of mountain road information.