Assessment of the Impact of Floods using Image Data acquired from a Helicopter

Publication from Digital

Raggam J., Wack R., Gutjahr K.H.

26th Earsel Symposium, “New Deveopments and Challenges in Remote Sensing”, Warsaw, May 29th – June 1st , 2006


In summer 2005 many heavy rainfall events were observed over mid-Europe, causing partly severe flooding of certain areas. An area in the east of Austria was extremely affected by flooding and consequent landslides. In order to assess the damage more clearly, multiple over-flights were made by a helicopter and overlapping images were acquired, using a high resolution digital consumer camera and interactive photo shooting. In parallel, the positions of the helicopter and the positions of the camera, respectively, were measured by DGPS with respect to each individual photo acquisition. The ground resolution of the images was in the range of slightly more than 1 decimetre. A high overlap was aspired in order to achieve full stereo coverage and 3D mapping capability. First, these data should be used for rapid mapping. A fast delivery ortho-image mosaic was to be generated in order to get a quick overview on the amount of landslide damage. Aero-triangulation was applied in order to determine the exterior orientation of the images, using a couple of lower quality control points, which were collected from ortho-photo maps existing at a pixel size of 50 centimetres, and a major set of tie-points. An available reference DEM with a baseline mesh width of 10 meters was then used for the ortho-rectification of the images, leading to products of reduced geolocation accuracy. Second, these lower-quality ortho-images, which still are affected by displacements due to reference DEM deficiencies, were used to reconstruct the terrain surface. Therefore, the displacements of the ortho-images in the individual overlap areas were determined by image matching, and these matching results were used to determine associated height increments. These use to refer to the height errors inherent to the reference DEM on the one hand, and to terrain cover, which is not included in the reference DEM (like trees and buildings) on the other. Thereby, a surface model could be generated from the fast delivery ortho-images. In the paper, first results which were achieved in this experiment are presented. The work is still in continuation, and covers related investigations and algorithm improvement in order to evaluate ultimate GCP requirements for such scenarios, to achieve superior surface models and to generate true ortho images.