Ground filtering and DTM generation from DSM data using probabilistic voting and segmentation

Automated digital terrain model (DTM) generation from remotely sensed data has gained wide application areas due to increased sensor resolution. In this study, a novel ground filtering and segmentation method is proposed for digital surface model (DSM) data. The proposed method starts with extracting DSM feature points. These are used in a probabilistic framework to generate a non-ground object probability map in spatial domain. Modes of this map are used as seed points in a novel segmentation method based on morphological operations. This leads to ground filtering and DTM generation. The method is tested on three different data sets. Two of these originate from light detection and ranging (lidar) sensors, where resulting kappa coefficient (kappa) range mostly higher than 95% for differently structured urban areas. Also, the visual appearance of the generated DTM exhibits obvious improvements over all other investigated methods. The third data set is a DSM obtained from WorldView-2 stereo image pairs. Also here, we compare our results with three different methods in the literature. Although the DSM quality is much lower, more than 85% of kappa can be reached by the proposed method, showing its superiority over other methods. Overall experimental results show that the proposed method can be used reliably for DTM generation. The results also indicate that the method has prominent advantages in comparison to established methodologies in terms of robustness in handling urban areas of different properties. Moreover, there are only few parameters to adjust in the proposed method, and these are independent of the object size in DSM data.