A rigorous model for high resolution satellite imagery orientation

The High Resolution Satellite Imagery -HRSI- has relevant impact for cartographic applications, it proved to be a suitable alternative to aerial photogrammetric data to produce maps at 1:5000 scale or lower, and for the generation of Digital Elevation Models (DEM). However, the real possibility of using high resolution images for cartography depends on several factors: sensor characteristics (geometric and radiometric resolution), types of products commercialized by the companies managing the satellites, cost and time to obtain these products, cost of commercial software for processing such products. The distortions sources can be related to two general categories: the acquisition system, which includes the platform orientation and movement and the imaging sensor optical geometric characteristics, and the observed object, which takes into account atmosphere refraction and terrain morphology. At present, orthorectification methods for the high resolution satellite imagery can be classified in three categories: black-box models (like Rational Polynomial Function -RPF), which consist in purely analytic methodologies, independent of specific platform or sensor characteristics and acquisition geometry; physically based models (rigorous model), which take into account several aspects influencing the acquisition procedure; the gray models (RPC - Rational Polynomial Coefficients models), in which RPF are used with known coefficients supplied in the imagery metadata and “blind” produced by companies managing sensors by their own rigorous models. The RPC approach results may be enhanced by users estimating some additional parameters (e.g. an affine transform) on the basis of eventually available ground control point (GCP). The accuracy of the final product strictly depends on the original image characteristics, on the quality of the known coordinates of ground control points (often obtained by GPS surveys) and on the model chosen to perform the orthorectification; in this respect it is well known that the RPF model is more inadequate for the geometric correction of satellite imagery for cartographic application than the rigorous photogrammetric approach; RPC represent a good compromise for not specialized users. Since 2003, the research group at DITS (Geodesy and Geomatics Area) has been developing and implementing a specific and rigorous model for the orientation of EROS A and QuickBird satellite imagery in a C++ software (SISAR - Software Immagini Satellitari ad Alta Risoluzione). The main results of this work are outlined in this thesis, whose main objectives are: to define the features of the model, to implement the model in the SISAR software, to validate the SISAR results and to define an effective methodology to assess the spatial accuracy achievable from oriented (and orthorectified) imagery. To point out the effectiveness of the new model, SISAR results are compared with the corresponding ones obtained by the software OrthoEngine 9.1 (PCI Geomatica), where Thierry Toutin’s rigorous model for the imagery elaboration of the main high-resolution sensors is implemented, and by Sat_PP software implemented at ETH Zurich in the research group of Prof. Gruen. The rigorous models were tested on 8 images with different features, for each image the model intrinsic precision and the image accuracy are studied considering Root Mean Squared Error (RMSE) of GCP and CP residuals. The analysis of results underline that Sat_PP and SISAR show a similar RMSE trend for all images, while OrthoEngine shows some unexpected behavior: it differs from the other two software with respect to blunders. The accuracy determination of a satellite images allows to state its real potentiality; for this reason we have studied and implement two methodologies for the accuracy evaluation: the Hold-Out Validation (HOV) and the Leave-One-Out Cross-Validation (LOOCV) methods. The most common method, the HOV method, consists in partitioning the known ground points in two sets, the first used in the orientation-orthorectification model (GCPs- Ground Control Points) and the second to validate the model itself (CPs – Check Points); the accuracy is the RMSE of residuals between imagery derived coordinates and CPs coordinates. An alternative to the previously described method to perform a spatial accuracy assessment is the use of the LOOCV method for the orientation (and orthorectification) of HRSI. The method consists in the iterative application of the orientation model using all the known ground points (or a subset of them) as GCPs except one, different in each iteration, used as CP. In every iteration the residual between imagery derived coordinates and CP coordinates (prediction error of the model on CP coordinates) is calculated; the overall spatial accuracy achievable from the orthorectified image may be estimated by calculating the usual RMSE or, better, a robust accuracy index like the mAD (median Absolute Deviation) of the prediction errors on all the iterations. The methods for accuracy evaluation have been applied on four images: two EROS-A1 and two QuickBird Basic images. The LOOCV accuracy evaluated by means of the robust index mAD is globally closer to the median of the HOV RMSEs with (n-1) GCPs and also to the HOV RMSE assumed as a reference than LOOCV RMSE. Therefore the LOOCV method with accuracy evaluated by mAD seems promising and useful for practical cases. Anyway, some critical cases were evidenced, mainly in connection with ground points placed on the perimeter of the areas covered by them, when they act as CPs. It has to be supposed that bad results stemming from these cases are not representative of the sought mean orientation accuracy, so that they have to be filtered out by a robust accuracy index. Finally, when there are one o more outliers in the model the accuracy variations and the behaviors of two rigorous models (OrthoEngine and SISAR) are analyzed.