Orientation models of optical High Resolution Satellite Imagery: definition, implementation and validation of original algorithms

High resolution satellite imagery became available to civil users in 1999 with the launch of Ikonos, the first civil satellite offering a spatial resolution of 1 m. Since then other high resolution satellites have been launched, among which there are EROS-A (1.8 m), QuickBird (0.61 m), Orbview-3 (1 m), EROS-B (0.7 m), Worldview-1 (0.5 m) and GeoEye-1 (0.41 m), with many others being planned to launch in the near future. High resolution satellite imagery is now available in different formats and processing levels at an affordable price, so that they already represent a possible alternative to aerial imagery, for cartographic applications and orthophoto production, especially for areas where the organization of photogrammetric surveying may be critical. Moreover, an increasing demand for terrain modelling exists so that almost all the satellites have along-track stereo acquisition capability. Many new satellites dedicated to stereo viewing, for example Cartosat-1 (2.5 m), have been launched. This enables the generation of Digital Elevation Models (DEMs) and Digital Surface Models (DSMs), and also for 3D feature extraction (e.g. for city modelling). The geomatic utilizations of satellite imagery for cartographic applications and terrain modelling require a high level geometric correction through image orientation. Some fundamental features related to the sensor models and their parameters estimation, both for single images and stereopairs orientation, were addressed and some real applications were discussed. In details, they were concerned both physical sensor models (also called rigorous models) and generalized sensor models (also called RPC models) for the orientation of basic images (level 1A) and of the image projected onto a specific object surface (usually an expanded ellipsoid derived from the WGS84) (level 1B). As regards the rigorous models, a thorough investigation of the fundamentals of their functional model was developed and the problem of parameters estimability was concerned, proposing a solution based on SVD and QR decomposition. RPC models were discussed not only with respect possible refinements by zero and first order transformations but also (and mainly) with respect the RPCs generation, based on previously established rigorous model; thanks to SVD and QR decomposition, it was showed that many RPCs are not estimable parameters, therefore they are not necessary to obtain the best achievable accuracy level. Real applications demonstrated that rigorous and RPC models both for Level 1A and Level 1B imagery can provide orientation accuracy at 1-1.5 pixel level in the horizontal components, and at 1-2 pixel level in the height for stereopairs (even better with Cartosat-1 and slightly worse with EROS-1).