University of Twente Student Theses
Object based image analysis to extract urban form from high resolution satellite image: A case study in Ahamabad City: India
Raihan, Imroz (2009) Object based image analysis to extract urban form from high resolution satellite image: A case study in Ahamabad City: India.
PDF
5MB |
Abstract: | Fast growing developing cities like Ahmadabad of India facing acute problems with urban sustainability especially with respect to urban transportation and the urban environment. Increasing pressure of urban transport triggers to deploy transport ecological footprints which have been considered as major determinants for monitoring negative impacts of this sector to environment. Urban form and urban land use information are intrinsically linked to travel pattern as well transport ecological footprints. Therefore the availability of urban form and urban land use information will make it possible to calculate the transport ecological footprints and its impacts. The urban form elements clearly are of interest to urban remote sensing component. This research aims at the application remote sensing to retrieve building footprints which will provide supports to develop urban form characteristics. However previous research proved the limitation of pixel based classification of remotely sensed images and unsuccessful to address urban form and land use information. Indeed manual photo interpretation of high resolution satellite image/aerial photography proves to be time consuming. The novelty of this research lies into experimentation of object based image analysis to extract building rooftops in the dense urban canyon using multi image. To extract building rooftops a test site of Ahmadabad City was selected which is located in a dense urban area with the area extent around 850X 850 meters. Three different satellite images were used simultaneously to extract building footprints. Initially Cartosat Panchromatic and IRS multispectral images were merged together to produce images having better spectral properties. At later stage Google Earth mosaic image were used to extract building outlines by applying multi-resolution segmentation. The purpose of using Pan sharpened Cartosat image is to get better spectral properties which can be used to separate building rooftop from other land cover. There were also the uses of DSM to separate building rooftops form other land cover class at ground level, remarkably the DSM was generated from Cartosat stereo pair image. Before applying image segmentation and image fusion all data layers were geo-rectified by applying advanced point matching technology which generates precisely geo-rectified multiple image layers. Theatrically there is no established image segmentation algorithm available and most of the segmentation techniques are still experimental stage. Indeed it’s uncertain to know which segmentation algorithm can efficiently address building outlines. Therefore four different image segmentation software’s (Definiens, Spring, ASTRO, Porabat) were chosen to test their quality. The both qualitative and qualitative (stand alone/empirical discrepancy method) assessment was made for all the segmented results. Quantitative assessment like average difference of areas, parameter ii suggests Definiens and Spring results the best image segmentation result however data processing in SPRING is quite difficult and it does not have good feature extraction capability. Therefore Definiens segmentation results were considered for further analysis. In later stage advanced image segmentation method were applied to produce better segmentation result this includes putting weights on image layers or making selection of spectral band for better image segmentation. Image segmentation results were optimized by reducing the radiometric resolution form 8 bit to 4 bit data depth. Different classification method was followed to optimize the classification result. To make comparison between object and pixel based classification cheese board segmentation were applied and brightness threshold value were determined for building rooftops. As usual scenario the pixel classifier produces results like salt and pepper and object classifier resituated continuous polygon. In addition with buildings some objects at ground level having similar brightness were also selected with the buildings. Later Nearest Neighbourhod(NN) classification were tested by using several land cover class. Results form NN classification visually evaluated and observed that the desired building class were not correctly classified. To have better classification of building footprints Feature Space Optimization tool were applied which also failed to retrieve the appropriate building rooftops. Efficient knowledge based rule set were developed by using three variables which includes mean brightness length and elevation. Threshold values for the three image layers were sequentially determined by changing their properties and visually checking the results. Finally the classified results were evaluated by applying overlay analysis using reference polygon. By comparing the total area difference between the reference polygons and the building rooftops there were in total 90% area matching were estimated. Whereas considering positional accuracy there was only 65% areas of classified building falls inside the reference polygon. Large amount of commission error (81%) appeared due unexpected impervious surface at ground level adjacent to buildings were misclassified as buildings. Finally there were the discussions about the use of building rooftops to formulate urban form characteristics which can ultimately feed developing transport indicator. As most characteristics of urban from can only be measured at the city scale, however the building footprint derived in this research covers only the test site. Therefore none of the characterization was estimated in this research rather description discussions were provided which indicates how building footprints can be used to formulate urban form. |
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: | https://purl.utwente.nl/essays/92716 |
Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Repository Staff Only: item control page