DTM generation from KOMPSAT-3A stereo images by utilizing an existing low-resolution DTM

초록

The launch of CAS500-1/2(Compact Advanced Satellite 500) is scheduled in 2020. The generation of 3D topographic information such as Digital Surface Model (DSM) and Digital Terrain Model (DTM) using optical images derived from CAS500 is expected to be in high demand. Conventional algorithms for DTM generation try to remove non-ground points and find the ground points in DSM. It is difficult to find the correct bare-earth points in the DSM itself in case of the tree canopy and urban areas. In addition, these techniques are sensitive to the accuracy and precision of the data. Therefore, proper methods are needed for DSM acquired from optical images. We propose a method to extract non-ground points existing in DSM generated from stereo image matching by applying the low-resolution DTM that has been already built. The main concepts of DSM to DTM generation consist of non-ground point extraction, classification of building and tree canopy, and ground height estimation. Initial normalized DSM (nDSM) is created using the height difference between DSM and low-resolution DTM. As a result of performing geodesic dilation, which is offset of the initial nDSM, grid values above threshold are defined as non-ground area. Height variations are calculated along the contour of each non-ground area. The areas with a high proportion of height jump point are classified as buildings and the rest as tree canopy. In areas classified as tree canopy, the height value of DSM is decreased by offset derived from the nDSM. In contrast, in the building area, the surface value is removed, and the ground height is estimated by Inverse Distance Weighted (IDW) interpolation. In the experiment, we obtained DSMs with 2 meters of Ground Sampling Distance (GSD) from the KOMPSAT-3A stereo images by using photogrammetric image matching technique. Shuttle Radar Topography Mission (SRTM), which resolution is about 30 meters, was used as the low-resolution DTM. The target areas were selected with four challenging locations that reflect various terrain characteristics. The proposed method is evaluated against a reference DTM, which is manual data provided by National Geographic Information Institute (NGII) of South Korea. Our method flexibly extracts non-ground points that exist in various regions and generates improved results compared to the SRTM. © 2020 40th Asian Conference on Remote Sensing, ACRS 2019: Progress of Remote Sensing Technology for Smart Future. All rights reserved.

키워드