Abstract: For extracting ground points from LIDAR points clouds quickly and generating high accurate DTMs effectively,a filtering method based on multi-resolution directional prediction is proposed.The method works by constructing multi-resolution datasets,smoothing datasets from low to high resolution based on directional prediction,and comparing the original LIDAR points clouds with the smoothed dataset to get bare ground points.The method improves the algorithm of "Inverse Distance to a Power" based on fixing its deficiencies in interpolating ground points.Highly accurate DTMs are interpolated by the proposed algorithm using bare ground points.It is confirmed that the algorithm is efficient in removing non-terrain objects and preserving terrain features,and holds promising potential for applications.