Urban Sprawl |
Fringe Area Development |
Urban Agglomeration |
Emerging Technologies |
Event-Driven Change-Detection in Urban Environments Using SAR
University of Stuttgart / Institute for Photogrammetry
SAR sensors are able to operate under nearly all weather conditions, at daylight or in the night. This is especially beneficial for event-driven applications, like time-critical change-detection for disaster management applications. Unfortunately SAR systems suffer from occlusions and ambiguities, especially in urban areas. Additionally due to layover and foreshortening effects the geo-referencing of SAR images, which is a prerequisite for a change detection, is problematic in urban areas. These effects can be reduced by event-driven SAR data acquisition based on SAR simulations.
Geo-referencing the data is crucial for change-detection. Referencing the SAR data automatically to street vectors allows a fast geo-referencing. For this purpose, the street vectors should be transformed to represent the SAR imaging properties. By comparing image chips, from the SAR image, to the transformed street data, correspondences can be found. These correspondences can be used for geo-referencing the SAR data.
For disaster management in urban areas the fast detection of collapsed buildings is most important. Therefore, different datasets have to be compared, to analyse changes in buildings. The original dataset, showing the status before the disaster took place is normally of a different type, than the data acquired directly after the disastrous event. Therefore the change-detection requires the fusion of different types of data, normally acquired by different sensors.
Change-detection with SAR images should be based on 3D-data, due to the side-looking property of SAR images. If no actual 3D-data is available, 2D data should be extended to the third dimension, i.e. using GIS-operations. The 3D-data is afterwards SAR simulated and the simulated SAR image is compared to the real SAR image. This comparison reveals changes in the building shapes, which can be used for detecting collapsed buildings.
Unfortunately some false alarms may still occur due to wrong models, occlusions, side lobes from corner reflection, or other problems. The influence of corner reflections can be reduced by filtering. It can be shown that using even simple models and assumptions in the simulation, like for example lambertian reflection, the results are quite good. The idea will be demonstrated using different city models and DOSAR images.