Comparison of DSM Generation Methods of Urban Areas from Ikonos Images

Abstract

In this paper two methods, an area based matching and a so called dynamic line warping, for generating digital surface models (DSMs) from IKONOS image pairs are compared and their limitations and usability for the rapid generation of good DSMs are discussed. The new dynamic line warping method is based on techniques used in speech recognition for warping corresponding epipolar lines of the stereo images onto each other in order to extract so parallaxes and object heights. One critical parameter for the derivation of DSMs is the convergence angle of the two stereo images. A small convergence angle gives similar images with smaller parallaxes and is better for area based matching techniques but is said to result in larger height errors and lacks of views from vertical features like walls. A large convergence angle ­ as used in the standard stereo products of most satellite image providers ­ on the other hand leads in urban areas to many features hidden in one or the other image and very different viewing conditions leading to more problems for intensity based matching. Therefore three IKONOS scenes with different convergence angles are used and the results are compared qualitatively.