High-level representation of 3D models of buildings

The goal of the current research was to design and implement a system able to extract high-level representations of 3D building models for the purpose of developing cognitive models. We present an intermediate representation scheme that supports modeling at different levels of detail. A cognitive model could easily use our representation scheme to perceive and navigate a 3D building. The overall goal of the project was to develop an implementation-independent representation scheme, sufficient to support high-level symbolic cognitive models. Our node-based representation supports high-level visual perception (which can be used in a spatial representation internal to the agent) or a complete semantically labelled topological map. Our representation scheme identifies structural features of the 3D building environments: “floor”, “ceiling”, “wall”, “door”, and “corner”, which are useful for agents programmed in symbolic cognitive architectures and navigating simulated environments. Our evaluation shows that this representation is sufficient for agents to do simple navigation in virtual environments.