The absence of curbs on national and provincial highways makes it challenging to identify road edges directly from point cloud data,thereby hindering accurate road point cloud extraction. Here,we propose a novel extraction method that combines shape feature vector screening with local roughness analysis. First,a grid index is constructed for the original point cloud data. Using each grid cell as a basic unit,we design a covariance matrix-based method to analyze shape feature vectors. This method calculates the vectors of the points within a cell,constructs an orthogonal basis to compute eigenvalues,and subsequently filters to obtain the ground point cloud. Second,a KD-tree index is used to reconstruct the ground point cloud,which facilitates the calculation and analysis of local roughness. By applying a pre-defined roughness threshold and employing a Euclidean clustering algorithm,the point clouds of the road surface and its immediate surroundings are extracted,effectively separating road points from non-road points. Tests on mobile laser scanning data covering 10 km of national and provincial trunk roads show that our method achieves an average precision of 99.64% and a recall of 99.59%. Compared to the standard Euclidean clustering segmentation algorithm,it improves average precision and F₁-score by 8.95 and 4.57 percentage points,respectively. In terms of computational efficiency,the algorithm processes original point cloud data at a speed of 32.4 km/h,showing a significant efficiency gain while maintaining high accuracy and robustness for road surface extraction.