Abstract:

Constructing 3D geospatial terrain databases for live, virtual, and constructive simulation applications is costly and time consuming.  To support the US Army’s Synthetic Training Environment (STE) terrain database production requirements many new automated processes have been developed to lower costs and reduce timelines. One of these automated processes is procedural 3D building model generation.

 Creating a procedural 3D building model generation capability for virtual simulation applications poses many complex challenges.  Four main requirements were identified for the creation of 3D buildings: 1) ingest a variety of source data types and input formats, 2) handle missing or incomplete building descriptions, 3) generate 3D building geometry with desired appearance for both exteriors and interiors, 4) fulfill a broad range of fidelity and functional requirements for the runtime simulation systems.

This paper describes a methodology for procedurally creating 3D building models using a centralized floorplan-based common source input into a 3D building model generation process. This approach, 3D Buildings from Floorplans (3DBFF), utilizes a floorplan-based schema for defining a building in a common Drawing Exchange Format (DXF). This floorplan interface is assembled from a variety of disparate data sources including from real-world reference data such as building footprints, CAD floorplans, building blueprints, satellite captured imagery, and drone captured imagery; and can be generated from manual, procedural, machine learning, and generative AI processes. The paper will describe why the DXF format was chosen as the central format type, including advantages from CAD software support, metadata structures, and common workflow. The paper will also describe the processes to transform this common data format into a fully functional 3D model of real-world buildings, with functioning windows and doors, multistory structures and traversable stairs and ladders. Last, we will show the application of this approach in multiple use cases, and enumerate our successes, failures, and lessons learned.

Keywords: MACHINE LEARNING;MODELING;SYNTHETIC ENVIRONMENT