University of Twente Student Theses


Improving the Quality of Digital Surface Model Generated from Very High Resolution Satellite Stereo Imagery by Using Object Oriented Image Analysis Technique

Joshi, Janak Raj (2010) Improving the Quality of Digital Surface Model Generated from Very High Resolution Satellite Stereo Imagery by Using Object Oriented Image Analysis Technique.

[img] PDF
Abstract:Digital Surface Models (DSMs) are widely used in many applications such as urban planning, disaster monitoring, surveying and mapping, construction and development. Very high resolution satellite (VHRS) image is one of many sources for DSM generation. Main advantage of VHRS image is its versatile application in geo information acquisition. Image matching techniques enable to extract digital surface models fully automatically, whereas high spatial resolution supports the extraction of urban features such as individual buildings. Many matching techniques in VHRS images are currently in use, but most of them do not produce anticipated result especially to preserve the building shape and outlines. Similarly (semi)automatic extraction of individual buildings, even in the highest resolution commercial data, is very difficult in a heterogeneous dense urban environment. Objectoriented analysis (OOA) methods have been used successfully in the image processing of some builtup area type, but success is highly dependent on settlement density, contrast and image characteristics. In this research, a stereo pair of VHRS images (GeoEye-1) of Cairo is used. This research explores a possibility of using DSM generation and OOA technique together in such a way that one can benefit the other. To extract buildings accurately from VHRS images, a DSM can play an important role, whereas the building outlines extracted from the image by OOA can, in turn, help refine the quality of DSM. This mutual relationship between DSM and OOA can benefit each other and using these two techniques together can help optimally use the VHRS image. The research was carried out in three main stages. Firstly, different aspects of DSM generation were explored to extract a DSM and produce an orthoimage. Then in a second step, the orthoimage was used in OOA for extracting building outline, in which the DSM was used as additional ancillary input data. Finally in the third step, the building outlines were used to improve the quality of DSM generated in the first step. The result shows DSM generated from VHRS image by matching technique plays an important role in building extraction by OOA. In relatively open areas, 98% of the buildings are extracted within a positional accuracy of 1.1 m. The accuracy in terms of correctness and completeness of building extraction is more than 90% in sparse building areas and that around 85% in dense areas. Similarly, the quality of building DSM in urban area is also improved significantly. Most of the problems and artefacts in the photogrammetric DSMs like oversmoothing bell shape, spikes and streaking are removed. The output DSM, after improvement, models the narrow streets and gap between the buildings. The DSM preserves building outline and edges. This method also removes the bell shape and streaking effect in DSM. The accuracy of the improved DSM in terms of shape and position, however, depends upon the accuracy of the building outline and the height depends upon the accuracy of input DSM used. Keywords: digital surface model, orthoimage, very high resolution satellite, object- oriented analysis, building extraction
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page