Comparison of filter algorithms and combination analysis for DEM extracting based on airborne laser scanning point clouds

Light detection and ranging (LiDAR) solves the shortcomings of traditional measurement technology, and becomes a new method to obtain digital elevation model (DEM). For forest areas in different terrains, it is a key step to extract the forest topography by choosing a reasonable point cloud filtering algorithm. In this paper, three blocks representing gently slope area, steep area and complicated area were selected. Three filtering algorithms were adopted in processing laser scanning point clouds, and the applicability of these three methods in different terrains was evaluated. Compared with the DEM derived from 1:10 000 topographic map, the results showed that different algorithms had varied adaptive regions. All of these three algorithms had good accuracies in gently slope area, the R² was up to 0.98 and RMSE was lower than 0.21 m. In complicated area, the iterative linear least squares prediction had the best performance, R² was 0.94 and RMSE was 0.21 m. In steep area, the algorithm based on triangulated irregular network (TIN) method had the best performance, R² was 0.99 and RMSE was 1.43 m. However, there were obvious classification errors when extracting DEM by single filtering algorithm in steep area or complicated area since vegetation could be divided into ground points. In order to improve the accuracy of DEM extraction under the forest, this paper proposes the combinations of different filtering algorithms. It was proved that the combination of iterative linear least squares prediction and TIN algorithms could obtain best result under the condition of reducing parameter adjustment.