Extraction of digital elevation model using stereo matching with slope-adaptive patch transformation

Traditional automated DEM generation from high-resolution satellite imagery relies on area-based correlation matching methods. However, these methods often fail, especially where the slope of the terrain is very high; this results in large geometric difference in the stereo pair. To address this issue, we introduced an adaptive stereo matching scheme that incorporates image transformation for the slope in the epipolar image space. The matching patch is designed to adaptively transform the target image according to the terrain slope determined in the lower scale of the image pyramid. This transformation reduces the geometric difference between the stereoscopic images due to the image acquisition angle and the terrain slope. Experimental results with Kompsat-2 images showed that the proposed method increased the overall matching success rate by 5% and the vertical accuracy of the generated DEM improved by about 2 meters (17%).