SEHAG - Historic image processing (SEHAG)
Remote sensing is an established method to monitor ongoing processes in the geosystem. Retrieving information on the past requires historic documents, therefore, old photographs are used extensively in cultural heritage studies. For documenting change in the alpine landscape photos have largely been used qualitatively, with some exceptions of aerial photography for quantitative evaluation, e.g. of glacier extent and height. These approaches rely on a small number of images and the exterior orientation is determined in an only semiautomatic process, involving manual measurements. A fast, automatic, and reliable approach for the georeferencing of terrestrial historic photographs has so far not been investigated.
The (DFG/FWF) SEHAG research group aims to quantify changes in the geosystem since 1850, analyzed in time slices starting 1850, 1920, and 1980 up to now. Three alpine valleys with different elevation, exposition, and touristic development stages will be investigated: Kaunertal and Horlachtal in Austria, and Martelltal in Italy. Together with researchers from botany, geography, geology, hydrology, and meteorology the aim is to build a time series of photos, starting with historic terrestrial images from 1850, including aerial images from roughly 1950, and ending with the newest airborne laser scanning point clouds. These data sets will allow a quantitative analysis of the development of feature lines in the geosystem, including rock fall deposit boundary lines, snow extent on glaciers, distribution of vegetation patches, tree line, etc. The development of these features will then be related – if possible – to climate changes.
The aims of this subproject are to i) retrieve photos from archives and make them accessible to the research group, including the design of a metadata catalogue, ii) develop an automatic method to orient the images automatically, given the valley they were acquired in, iii) reconstruct relevant feature lines including quality measures, iv) orient image sets and derive a 3D terrain model for historic photos taken within a short time spans, v) develop terrain model and ortho photo time series from 1950 onwards using regular photo flights, and vi) integrate all data in a stable reference frame.
The automatic determination of the exterior and interior orientation of terrestrial images is based on the hypothesis, that the silhouette is unique within a valley. Using the silhouette as a line, but also a combinatorial approach using peak and saddle points (in the photo and from the terrain), will be investigated for the automatic georeferencing of the image. Monoplotting is the method for single image 3D reconstruction, exploiting the best available terrain model for the respective time. Computing and communicating accuracy figures for these reconstructions is highly important, as only significant changes shall be considered. Finally, for the stable reference frame current airborne laser scanning will be used.