OpenAIRE is about to release its new face with lots of new content and services.
During September, you may notice downtime in services, while some functionalities (e.g. user registration, login, validation, claiming) will be temporarily disabled.
We apologize for the inconvenience, please stay tuned!
For further information please contact helpdesk[at]openaire.eu

fbtwitterlinkedinvimeoflicker grey 14rssslideshare1
Oczipka, Martin Eckhard (2010)
Publisher: Freie Universität Berlin Universitätsbibliothek, Garystr. 39, 14195 Berlin
Languages: German
Types: Doctoral thesis
Subjects: HRSC obejct based classification LIDAR DSM, 550 Earth sciences, 550 Geowissenschaften
ddc: ddc:550

Classified by OpenAIRE into

ACM Ref: ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION
Over the last couple of years more and more analogue airborne cameras were replaced by digital cameras. Digitally recorded image data have significant advantages to film based data. Digital aerial photographs have a much better radiometric resolution. Image information can be acquired in shaded areas too. This information is essential for a stable and continuous classification, because no data or unclassified areas should be as small as possible. Considering this technological progress, one of the basic questions is how the potential of high radiometric and geometric resolution data can be used in an automatic analysis particularly in urban regions. For this study an object-based classification algorithm was selected to evaluate its suitability to update maps. Pixel-based classification algorithms are problematic for the classification of high resolution image data, as the contiguous objects often are separated due to their spectral variability. Object based classification algorithms are a good alternative due to their ability to create objects which represent semantic objects. In this thesis, image data of the digital sensor High Resolution Stereo Camera - Airborne eXtended, HRSC-AX, an extended version of the HRSC-A, was used. The construction follows the concept of a pushbroom scanner. Within the photogrammetric processing four multispectral bands, red, green, blue, near infrared, as well as five panchromatic bands are used to create true orthophotos and a digital surface model (DSM). Typically the geometric resolution is 20 cm in X, Y and a decimetre in Z. The image data was not converted from 12bit to 8bit in order to prevent loss of information. In this study HRSC-AX image and DSM data from Berlin was used to develop and test an automated classification procedure in the commercial software Definiens Developer. Alternatively a LIDAR-DSM was used in the segmentation process. A comparison and accuracy assessment of both data sets was done evaluate their suitability for the segmentation process. Comparing the standard deviation of the DSMs to ground control points the LIDAR-DSM proved to be more reliable. Both data sets were tested in the image analysis algorithm. DSMs are essential for the separation of vegetation classes and different buildings and can stabilize the classification result. Advanced software, like Definiens Developer allows, transferring the process tree of one analysis to different data sets of the same sensor and to adapt the algorithm to other sensors and conditions. Comparing image analysis results using both DSM surprisingly show very little differences. This is connected to the object-based classification grouping pixels to objects and with doing so, smoothes out errors. The overall Kappa statistics for the classification was 0,8709 for the image analysis process using multispectral data HRSC and DSM and 0,8646 using multispectral data and a LIDAR-DSM. The image analysis process without using any DSM in the segmentation shows an overall Kappa of 0,8708. Besides the very small differences in the static, the visual evaluation of segmentation results leaves the combination of HRSC multispectral data and LIDAR DSM to be the most promising. Further qualitative analysis was executed using the German cadastral geographic information system (ALK). The intersection of classification results and the ALK disclosed errors and problems in updating the system. A direct update of the system is usually not possible because the ALK is based on the plan of the buildings, a feature which is generally difficult to identify in an orthophoto. Still, creating an intersection of the cadastral data and the classification results help to detect changes in build up areas and vegetation. Intersection of classification results and the ALK can be used to monitor impervious surface within a block. Although automatic updating of maps is not possible, image analysis eases monitoring.
  • No references.
  • No related research data.
  • No similar publications.

Share - Bookmark

Cite this article

Cookies make it easier for us to provide you with our services. With the usage of our services you permit us to use cookies.
More information Ok