MORPHOLOGICAL FILLING OF DIGITAL ELEVATION MODELS

Abstract. In this paper a new approach for a more detailed post processing and filling of digital elevation models (DEMs) in urban areas is presented. To reach the required specifications in a first step the errors in digital surface models (DSMs) generated by dense stereo algorithms are analyzed and methods for detection and classification of the different types of errors are implemented. Subsequently the classified erroneous areas are handled in separate manner to eliminate outliers and fill the DSM properly. The errors which can be detected in DSMs range from outliers – single pixels or small areas containing extremely high or low values – over noise from mismatches, single small holes to occlusions, where large areas are not visible in one of the images of the stereo pair. To validate the presented method artificial DSMs are generated and superimposed with all different kinds of described errors like noise (small holes cut in), outliers (small areas moved up/down), occlusions (larger areas beneath steep walls) and so on. The method is subsequently applied to the artificial DSMs and the resulting filled DSMs are compared to the original artificial DSMs without the introduced errors. Also the method is applied to stereo satellite generated DSMs from the ISPRS Comission 1 WG4 benchmark dataset and the results are checked with the also provided first pulse laser DSM data. Finally the results are discussed, strengths and weaknesses of the approach are shown and suggestions for application and optimization are given.