Stereo Based Very High Resolution Satellite Image Classification Using Rpcs

Detection of urban objects in very high resolution (VHR) satellite imagery is challenging due to the similarities in the spectral and textural characteristics of urban land cover classes. Therefore, additional information such as elevation data is required for a proper classification. In this study, instead of LiDAR data, elevation information generated from satellite stereo images is used to assist the urban land cover classification in VHR imagery. RPCs are used to generate the elevation information. The classification process is performed using a fuzzy inference rule based system. This method is tested on GeoEye-1 and WorldView-2 satellite imagery. Preliminary results suggest that urban land cover classification is substantially improved by adopting elevation information from the stereo imagery, after it is transferred to the image domain.