Traditional digital elevation model (DEM) is extracted from a pair of stereo images with a wide baseline. But wide baselines may increase the likelihood of occlusion problems and the difficulties of image matching. In this work, a method that generates DEM from multispectral images with very narrow baseline is demonstrated. A dense match algorithm based on 2-D fitting Phase Correlation (PC) method is applied to accurate estimate disparity maps of multispectral images. The experimental results using ASTER imagery show that DEM generated from multispectral images is feasible and the image pair is much easier for matching. The image matching accuracy is higher than 1/10th pixel in mountainous areas after error points eliminated. Comparing with the DEM of conventional baseline, the DEM from multispectral images has an acceptable accuracy. DEM generated from multispectral images is inferior to traditional DEM in details of mountain ranges. But the trends of the mountains and topographical features are presented clearly. And error analysis of this algorithm implies that the matching accuracy, distortion of sensors and satellite jitter could be the main factor which influence the quality of DEM extracted from multispectral images. © 2013. The authors.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below