3D geo-network for agent-based building evacuation simulation

This paper discusses 3D geometric network extraction for building evacuation simulation with an agent-based model. 3D geometric network represents the internal structure of a building, which provides agent-based models with the shortest path for evacuation. 3D geometric network of a building can be built from computer-aided design (CAD) file (vector) and scanned blueprint (raster) through wall extraction and 3D topology construction. We test two wall extraction methods: vector-based medial-axis transformation (MAT) and rasterbased thinning. For vector-based MAT, straight MAT is used to extract wall structure from wall polygon generated from CAD data. For raster-based thinning, boundary peeling thinning is used to extract wall structure from scanned blueprint. The extracted 3D geometric network is then used in an agent-based model for building evacuation simulation. In the evacuation simulation, human beings are considered to be the only moving agents. To model human behavior, we adopt a social force model to consider human-to-human and human-to-wall interactions during evacuation. We test simple evacuation scenario in a situation of jam by enforcing different numbers of people in three rooms. The results show that the average velocity increases continuously before jams, decreases during jams at doorways and outer exits, and eventually increases again as individuals escape the jams.