Masters Thesis Abstract: Daniel O’Donohue
Correcting Directly Georeferenced Thermal Imagery Using GIS Data
Imaging systems combined with photogrammetry processes have proven their ability to produce vast quantities of detailed geographic data (Cleve C et al. 2005). However, increased data availability and detail also increases the amount of resource required to maintain and process data. This has lead to research into the automation of image analysis techniques with the goal of relating image features to representations of real world objects stored in existing spatial data databases. Relating image features to database objects enables database update and enhanced image classification routines that incorporate object attribute data as well as spectral information. Current techniques rely heavily on the spatial alignment of geographic datasets, however, this alignment cannot always be assumed. For example, direct georeferencing automates the process of georeferenceing aerial imagery, but spatial accuracy may vary greatly throughout the dataset.
This research aims to develop a method of automatically correcting the misalignment of directly georeferenced thermal imagery by extracting image objects and matching them against representations of real world features in a GIS database. Directly georeferenced thermal imagery of Christchurch City and a polygon layer representing the spatial footprint of buildings within Christchurch are the main inputs. This software will incorporate open source geospatial libraries and the use of an expert system. Comparisons of image features and database object will be based on geometric attributes and topological relationships. Successful matches between image features and database objects allow the images to be spatial corrected.