An edge extraction method from LiDAR point cloud data based on rotation difference kernel estimation is proposed. For any point in the point cloud, the symmetrical center is the data point in a given direction, and the symmetrical window is constructed with a certain distance.The Kernel function about distance is defined in symmetric windows, and the weighted mean of elevations for the data points within the two windows is calculated. The absolute value of difference between the two weighted mean values is employed as edge magnitude of data point in the direction, and the maximum edge magnitude in all directions is selected as criterion for edge points. Then variances of elevations for the data points within the two windows in the direction corresponding to maximum edge magnitude is calculated, and the boundary points between buildings and ground are extracted by combining the absolute value of the difference between the two variances and the criterion of the edge points. By adjusting the distance between two windows, the maximum absolute value of the difference between the elevation variance in all directions is obtained, and this absolute value is used as the criterion of tree points. The absolute value of the difference between the two variances is used as the criterion of tree points, and the tree points are extracted after removing the junction between the building and the ground from the set of points detected by the criterion. The tree points in point cloud data are filtered by laser propagation characteristics, and then the complete building edges are extracted. The experimental results show that the proposed method effectively overcomes the influence of trees, and the accuracy of building edge extraction is about 80%.