3D Mapping with High Resolution Images

Abstract

High-resolution images, EROS-A1, IKONOS and QuickBird-2 from 2 m to 0.6 m pixel spacing re-spectively, are geometrically processed with a 3D parametric model developed at the Canada Centre for Remote Sensing. A positioning accuracy of one pixel for the ortho-images can be obtained if 7-10 ground control points (GCPs) used in the 3D parametric model computation are better than 1-pixel accurate (carto-graphic and image coordinates) and if the digital terrain model (DTM) used in the ortho-rectification proc-ess is more accurate than 5-m. When the GCPs are less accurate (around 3-5 m) 20 are necessary to avoid the error propagation through the 3D parametric model. Furthermore, a DTM is extracted from stereo IKONOS images using automatic image matching. A general accuracy of 6.5 m (68% level of confidence) when compared to an airborne lidar DTM (0.5 m accurate) is obtained but is correlated with the land covers. However, the accuracy on bare soils improves to 1.5 m (68% level of confidence). Since the sur-face heights are included in DTM, the surface and the height of buildings can be extracted from the stereo IKONOS DTM. Different methods are proposed, which should used an expert system to integrate a priori information on the buildings in the different processing steps.