Evaluation of Errors in Digital Terrain Models generated by High Resolution Satellite Images

Kurzfassung

Nowadays Digital terrain models play very crucial role in many applications, including engineering to design heavy construction project such as dams, tunnels and highways as well as orthophoto production and modeling and visualization in military applications. While new techniques such as LIDAR are available for almost instant Digital Surface Model generation, the use of stereoscopic high-resolution satellite imagery (HRSI), coupled with image matching, affords cost-effective measurement of surface topography over large coverage area. However automatic filtering algorithms should be used to extract the bare lands without vegetation canopy and buildings and classify surface to terrain and off terrains points. Additionally blunders may occur throughout DSM and DTM generation. At the department of photogrammetry and image processing at German Aerospace Center (DLR) a novel algorithm for automatic DTM generation from high resolution satellite images has been developed. It consists of two major steps: DSM generation and DTM generation. In the first step, Digital Surface Models (DSMs) are created from stereo scenes with emphasis on fully automated georeferencing based on semi-global matching. In the second step which is dedicated to DSM filtering, the DSM pixels are classified into ground and non-ground using the algorithm motivated from the gray-scale image reconstruction to suppress unwanted elevation pixels. In this method, non -ground regions, i.e., 3D objects are hierarchically separated from the ground regions. However this technique implies the risk of error and ill determined areas. The objectives of this thesis are to identify performance of the filtering algorithm and make a comparison with some others well known filtering algorithms and also type and magnitude of errors and corresponding contributions in generated DSM to mitigate the errors and outliers as much as possible. Additionally A method based on robust statistical estimation is presented to detect gross errors in DTMs. In the end it is concluded that general performance of filter algorithm is quite well in particular for vegetation areas. However, some difficulties in filtering are observed in complex landscape especially those that located on steep slopes. In the case of DSM generator algorithm computed accuracy respect to LASER data sets for region with hilly grass property is poor. Conversely it is observed that corresponding accuracy for DSM generated from area with residential and hilly bared characteristics follow the accuracy of LIDAR datasets very well.