Precise models of the impact of explosions in urban environments provide novel and valuable information in disaster management for developing precautionary, preventive and mitigating measures. In this project, we develop methods enabling accurate predictions of the process and effect of detonations at particular locations in urban environments.

Once an increased threat-level for a specific city is imminent, local disaster management units face three fundamental challenges:

1. Precaution: how to realistically simulate and train for a specific incident? What actions need to be taken at specific locations (evacuation, blocking, guarding, etc.).
2. Prevention: how to prevent a threat, or minimize its impact? Which locations are especially vulnerable and need protection?
3. Handling: is an actual threat accessible? How to evacuate people? What is the actual danger radius?

A proper reaction needs to be based on information which is recent, detailed, complete and easily accessible. More precisely, in order to perform a proper explosion simulation at arbitrary places, a complete city should be available as a semantic 3D model, at a sufficient level of accuracy. The 3D model should be fairly recent to resemble the current state of buildings, and include meta-information like surface material, vegetation and window locations. This is far beyond the standard information content of traditional GIS-databases.

Our project goal is to generate exactly this information from aerial images. Based on an automatically generated 3D city model we perform a semantic segmentation of the image data to detect object classes like roofs, windows, vegetation, etc. We fuse the geometric content and semantic content into a common representation to prepare the model for its use as geometry in numerical calculations and perform a detailed simulation of explosive expansion within the model.

Finally, we embed the 3D map and the simulation results into 3D visualization systems and Geo Information Systems (GIS) for use by disaster management units. For the first time, a photorealistic and intuitive visualization of threat scenarios will be available for training, planning and risk assessment.

This project has been funded by the Austrian Research Promotion Agency (FFG) under the KIRAS Program No. 860661