Models And Algorithms On Processing Of High-Resolution Satellite Remote Sensing Stereo Images

The extraction of spatial information from high-resolution remote sense image is a hot topic of digital photogrammetry, GIS and other fields. Aiming at the linear CCD push-broom of high resolution satellite imagery, the paper discusses theory and application on sensor geometry model, epipolarity model, and stereo image matching. At the end of the paper, the technique of extraction of regular building from high resolution remote sensing image is studied. The main work are as follows:1. the 2D affine projection model is analyzed from geometrical relation of parallel projection, the conversion of projection, the foundation of strict mathematical model of 2D affine. The characteristics of 2D affine projection are illustrated by experiments. The experiments show that this model can be used to derive orientation with high precision and robustness.2. As for the high resolution remote sense image, a general sensor model based on the rational function(RFM) is discussed and analyzed systematically from setting up of the model, algorithm of stereo intersection, control scheme and theoretical feature analysis. Results show that a properly modeled rational function can achieve an accuracy of rigorous model and can be used to complete photogrammetric processing in place of rigorous model, and, at the same time, the non-physical coefficients of rational function can realize hiding of the sensor's imaging information effectively.3. Through analyses of epipolarity geometry of the linear CCD push-broom images and frame perspective images, based on the projection track method, epipolarity of linear CCD push-broom image is established and the properties of epipolarity curves is discussed, and a new epipolarity model based on the simplified push-broom sensor model is proposed. For stereo images of frame perspective and linear perspective, how to set up of epipolarity is studied in terms of fundamental matrix in computer vision, and an algorithm of epipolarity solution that only depends on coordinates relation of identical point is presented.4. Through a deep study over the epipolarity based on the projection track method, the author puts forward the approximate line constraint method of dynamic epipolarity, and sets up constraint conditions of epipolarity in linear CCD push-broom stereo image matching, and proposes an imaging constraint method on the basis of application analysis of epipolarity, and taking imaging characteristics into account.5. Due to the remarkable difference in hue and the big geometric deformation caused by different imaging conditions in linear CCD push-broom images, a matching method is proposed that is based on the invariant moment of histograms, two-dimension invariant moment of images, and feature weighted gray value window. Not only does this method take gray information into account but it also make use of image feature. Compared with conventional methods, this method gains a better accuracy, efficiency and reliability.6. The technique of building extraction and reconstruction based on remote sense image is analyzed from primary data sources and cues, supported object models, extraction strategy, target detection and feature extraction, structuring, grouping, modeling and automatization. To tackle the problem of 3D regular building reconstruction from urban high-resolution remote sense imagery, an approach based on the 2D building detection technique and accurate 3D height information is present. Experiment results show that the method has better performance in the extraction of regular buildings in high-resolution remote sense images.