Radiometric and Geometric Evaluation of Ikonos Geo Images and Their Use for 3d Building Modelling

Investigations on the radiometric and geometric characteristics of IKONOS Geo satellite imagery and its use for orthoimage generation and 3D building reconstruction are reported. The paper first starts with an analysis of the radiometric quality of IKONOS Geo images of varying preprocessing and type, focussing on noise, edge quality and definition, and various artefacts. A noise estimation method is presented and results from various IKONOS Geo images, showing an intensity-dependent noise of 2-12 grey levels, which is high considering the effectively 8-9 bit data. Methods will be presented that reduce noise, enhance edges and contrast and optimally reduce 11-bit to 8-bit, without leading to loss of small, but still important, image details. Next, the topic of the geometric potential of IKONOS Geo images using the Rational Polynomial Coefficients (RPCs) and a possible improvement will be discussed. The geometric analysis makes use of stereo images and testfield data provided by the University of Melbourne with 28 well defined and distributed GCPs with 1-2 dm accuracy in object and image space. It will be shown that the RPCs have moderate absolute accuracy (large systematic bias, conforming to the Geo product accuracy specifications of 25 m RMS) but sub-metre relative accuracy. With the help of 4-8 accurate and well-distributed GCPs and a simple translation (bias removal), 0.4-0.5 m and 0.6-0.8 m absolute accuracies in planimetry and height can be achieved with the full set of 60 RPCs per image, or reduced RPCs omitting the higher-order terms (down to 22 coefficients). Even when using just one GCP to remove the bias, the accuracy is at worst ca. 2 m. In a related paper (Fraser et al., 2001b), it is shown that other simple sensor models (including affine projection) can produce similar or better accuracy than the RPCs with similar GCP requirements. This is shown also in this work, but with a different 3D spatial resection, using relief-corrected affine transformation and down to just 3 GCPs. Results from generation of orthoimages from three projects using own methods will be shown, including quantitative analysis, and showing how with simple and fast methods the 1-2 m accuracy of the much more expensive IKONOS Precision and Precision Plus can be achieved. With 1-2 dm accurate GCPs and depending on DTM accuracy and sensor elevation, sub-metre planimetric accuracy can be achieved. Buildings of the University of Melbourne Campus were measured manually and stereoscopically. After measuring a point cloud, 3D points were automatically structured and visualised with the system CC-Modeler, including texture mapping. The point accuracy was evaluated using GPS measurements of building corners and was found to be in the meter range. The completeness of the data and their limits regarding modelling roof details were evaluated: (a) qualitatively and (b) quantitatively through a comparison to a 3D model that was generated from aerial images. The results of this assessment are summarised, these highlighting both the high metric potential of IKONOS and difficulties to be anticipated in building reconstruction when the modelling should be complete and detailed.