Singularity Mitigation in Automated DEM Extraction from IKONOS Images

Extraction of a digital elevation model from high resolution satellite images such as IKONOS is important for many applications. One of the most popular techniques hired for automated DEM extraction is stereo matching. However, high resolution satellite images possess some properties that make automated stereo matching difficult. They present edges of individual buildings, building facades and shadows. The effects of height discontinuity and the presence of occluded regions are significant. They often possess very homogenous areas in urban regions and very noisy patterns on vegetated regions. We will first review the characteristics of high resolution images that make automated DEM generation difficult and analyzed their effects with one IKONOS stereo pair. We will discuss the ideas to mitigate these problems. We will show that the problems of occlusions and building facades can partly be solved by using two way stereo matching, once with the left image as the reference and once with the right image as the references and by merging the two results. We will also show that the problems of noisy patterns and building facades can partly be solved by using pyramid approach.